Hybrid Encoding Scheme for AMBTC Compressed Images Using Ternary Representation Technique

Abstract

Absolute moment block truncated coding (AMBTC) is a lossy image compression technique aiming at low computational cost, and has been widely studied. Previous studies have investigated the performance improvement of AMBTC; however, they often over describe the details of image blocks during encoding, causing an increase in bitrate. In this paper, we propose an efficient method to improve the compression performance by classifying image blocks into flat, smooth, and complex blocks according to their complexity. Flat blocks are encoded by their block means, while smooth blocks are encoded by a pair of adjusted quantized values and an index pointing to one of the k representative bitmaps. Complex blocks are encoded by three quantized values and a ternary map obtained by a clustering algorithm. Ternary indicators are used to specify the encoding cases. In our method, the details of most blocks can be retained without significantly increasing the bitrate. Experimental results show that, compared with prior works, the proposed method achieves higher image quality at a better compression ratio for all of the test images.

1. Introduction

With the rapid development of imaging technology, digital images are perhaps the most widely used media of the Internet. Because digital images themselves contain significant amounts of spatial redundancy, an efficient lossy image compression technique is required for lower storage requirement and faster transmission. The Joint Photographic Experts Group (JPEG) [1,2], vector quantization (VQ) [3,4], and block truncation coding (BTC) [5,6] are well-known lossy compression methods and have been extensively investigated in the literature. Among these techniques, BTC requires significantly less computation cost than others while offering acceptable image quality. BTC has been widely investigated in the disciplines of remote sensing and portable devices, in which computational costs are limited. BTC was firstly proposed by Delp and Mitchell [7]. This method partitions image into blocks, and each block is represented by two quantized values and a bitmap. Inspired by [7], Lema and Mitchel [8] propose a variant method called absolute moment block truncation coding (AMBTC), which offers a simpler computation than that of BTC.

The applications of AMBTC are studied in video compression [9], image authentication [10,11,12], and image steganography [13,14]. Moreover, some recoverable authentication methods adopt AMBTC codes as the recovery information to recover the tampered regions. Because the recovery codes have to be embedded into the host image, a more efficient coding of AMBTC is always desirable because the burden of the embedment can be reduced and the quality of the recovered regions can be enhanced. To improve the compression efficiency of the AMBTC method, several approaches, including bitmap omission [15], block classification [16,17], and quantized value adjustment [18], are adopted to lower the bitrate while maintaining the image quality. For example, Hu [15] recognizes that if the difference between two quantized values is smaller than a predefined threshold, the bitmap plays an insignificant role in reconstructed image quality. Therefore, Hu employs the bitmap omission approach by neglecting the recording of a bitmap if a block is considered to be flat, and only uses block means to represent the flat block. Chen et al. [17] adopt quadtree partitioning and propose a variable-rate AMBTC compression method for color images. The basic idea of [17] is to partition the image into blocks with various sizes according to their complexities. The AMBTC and bitmap omission technique are then employed to encode the image blocks. In some applications, such as data hiding or image authentication, bitmaps have to be altered to carry some required information, causing a degradation in image quality. Hong [18] optimizes the quantized values so that the impact of bitmap alteration can be reduced. Mathews and Nair [19] propose an adaptive AMBTC method based on edge quantization by considering human visual characteristics. This method separates image blocks into edge and non-edge blocks, and quantized values are calculated based on the edge information. Because the edge characteristics are considered, their method provides better image quality than other AMBTC variants.

Xiang et al. [16] in 2019 proposed a dynamic multi-grouping scheme for AMBTC focusing on improving the reconstructed image quality and reducing the bitrate. Their method partitions an image into non-overlapping blocks. According to the block complexity, varied grouping techniques are designed. An indicator is employed to distinguish the grouping types. In addition, instead of recording the quantized values, the differences between them are recorded so as to reduce the bitrate. Xiang et al.’s method provides better compression performance than those of prior works.

In Xiang et al.’s method, the number of pixel groups of an image block directly affects the reconstructed image quality and bitrate. Their method divides pixels of complex blocks into three or four groups during encoding, which may improve the image quality insignificantly but requires more bits for encoding. In this paper, we propose a ternary representation technique, which uses two thresholds to classify image blocks into three types, namely flat, smooth, and complex. We use the bitmap omission technique [15] to code flat blocks. The adjusted quantized values and an index pointing to one of the representative bitmaps are used to encode the smooth blocks. The complex blocks are encoded using three quantized values and a ternary bitmap. Compared with the AMBTC and Xiang et al.’s work, the proposed method achieves a higher reconstructed image quality with a smaller bitrate.

The reminder of this paper is organized as follows: Section 2 introduces AMBTC and Xiang et al.’s methods. Section 3 introduces the algorithms of this paper in detail. Section 4 presents the experimental results of the proposed method, and concluding remarks are provided in the final section.

2. Related Works In this section, we briefly introduce AMBTC and Xiang et al.’s methods, which are compared with the proposed method for evaluating the encoding performance. 2.1. The AMBTC Method

The AMBTC method [8] compresses image blocks into two quantized values and a bitmap. The detailed approaches are as follows. LetIbe the original image of sizew×hand partitionIinto non-overlapping blocks_{Ii}i=0N−1of sizen×n, whereN=(w/n)×(h/n)is the total number of blocks. Let_Ii,jbe thej-thpixel ofi-thblock. Therefore,_Ii=_{{Ii,j}j=0n×n−1}. For block_Ii, the averaged value_mican be calculated by:

_mi=1n×n∑j=0n×n−1_Ii,j.

Thej-thbit of bitmap_Bi, indicated by_Bi,j, is used to indicate the relationship between_Ii,jand_mi._Bi,jcan be obtained by:

_Bi,j={0,_Ii,j<_mi;1,_Ii,j≥_mi.

The lower quantized value_aiand higher quantized value_biare obtained by averaging the pixels in_Iiwith values smaller than and larger than or equal to_mi, respectively. This can be implemented by sequentially visiting pixels in_Ii. The lower quantized value_aiis obtained by calculating the averaged value of visited pixels with values smaller than_mi. Similarly, the higher quantized value_biis the averaged values of the other pixels. Therefore, the compressed code_Φiof_Iiis{_ai,_bi,_Bi}. Each block is processed using the same manner, and the AMBTC compressed codes_{{Φi}i=0N−1}=_{{ai,bi,Bi}i=0N−1}of imageIare then obtained.

To decode_{{Φi}i=0N−1}=_{{ai,bi,Bi}i=0N−1}, blocks_{{Ii′}i=0N−1}of sizen×nare prepared, where_Ii′=_{{Ii,j′}j=0n×n−1}. Thej-thpixel of_Ii′can be decoded by:

_Ii,j′={_ai,_Bi,j=0;_bi,_Bi,j=1.

After all of the image blocks are reconstructed, the image^I′can then be obtained.

2.2. Xiang et al.’s Method AMBTC uses the same approach to compress all image blocks. However, the same approach may not suitable for flat and complex blocks. As a result, Xiang et al. proposed an improved scheme to efficiently encode blocks according to their complexity, and achieve a better image quality than that of AMBTC with a satisfactory bitrate.

Let_{{Φi}i=0N−1}=_{{ai,bi,Bi}i=0N−1}be the AMBTC compressed code of the original imageI=_{Ii}i=0N−1. To determine the complexity of block_Ii, a threshold_τ0is set. If_bi−_ai≤_τ0, the variations of pixel values in block_Iiare relatively small. Therefore, all the pixels in this block are categorized as one group. In this case, the block mean_miis calculated, and this block is encoded by_(mi)2, which is the 8-bit binary representation of_mi.

If_bi−_ai>_τ0, the variations of pixels in block_Iiare large and these pixels need to be regrouped to achieve a better reconstructed image quality. Let_Gi0and_Gi1be the group of pixels with_Bi,j=0and_Bi,j=1, respectively. Apply the AMBTC method to_Gi0and_Gi1to obtain codes_Φi0={_ai0,_bi0,_Bi0}and_Φi1={_ai1,_bi1,_Bi1}. According to a given threshold_dmin, this method uses the following rules to determine whether_Gi0and_Gi1should be regrouped:

Rule 1: If_bi0−_ai0>_τ0and the total number of pixels in_Gi0is greater than_dmin.

Rule 2: If_bi1−_ai1>_τ0and the total number of pixels in_Gi1is greater than_dmin.

If neither rule is met, block_Iidoes not need to be further divided. Otherwise, block_Iiwill be sub-divided into three or four groups using the following rules:

(1)

If only rules 1 or 2 are met, group_Gi0or_Gi1needs to be subdivided. The number of pixels needing to be subdivided is denoted by_Pi, and_Pi-bitbitmap_Bi0or_Bi1has to be used to record the bitmap of_Gi0or_Gi1. In this case, block_Iiis eventually divided into three groups.

(2)

If both rules 1 and 2 are met, both_Gi0and_Gi1need to be subdivided, and block_Iiis eventually divided into four groups. Bitmap_Bi0and_Bi1have to be recorded to maintain the grouping information.

Xiang et al.’s method uses a 2-bit indicator_INDto record grouping information of_Ii. When block_Iiis divided into one to four groups, the indicator_INDis set to be₀₀₂,₀₁₂,₁₀₂, and₁₁₂, respectively. Moreover, if_Iineeds to be divided into three groups, an extra indicator is required to show which group is subdivided. Specifically, if_Gi0is sub-divided, then_Ji=0. On the contrary, if_Gi1is sub-divided, then_Ji=1.

To record the quantized values, Xiang et al.’s method records the smallest quantized value of a block using 8 bits, and utilizes a difference encoding scheme (DES) to encode the difference_dibetween two quantized values. In DES, if_di<γ, whereγis a predefined threshold,_diis recorded using_log2(γ)bits. Otherwise,_diis recorded using⌈_log2(σ)⌉bits, whereσis the maximum difference between quantized values in all blocks. An extra indicator_Yiis used to distinguish these two methods. That is, if_di<γ,_Yi=0is set. Otherwise,_Yi=1. The number of bits_Riused to record the difference can be expressed as:

_Ri={_log2(γ)+1,_di<γ;⌈_log2(σ)⌉+1,_di≥γ.

We use the symbol_(x−y)2to represent the R-bit encoded result of the difference betweenxandyusing DES. For example, ifx=40,y=28, andγ=64, then_di=12<γ. Therefore,R=7and the encoding result is_(40−28)2=₀₂||_0011002, where||is the concatenation operator.

The compressed code and the number of bits required to record blocks_Ii of different grouping cases are summarized in Table 1. Each block is compressed using the same procedures and the final compressed code streamC_Sfof imageIis obtained.

To decodeC_Sf, the 2-bit indicator_IND is read. According to the read bits, four possible compressed codes shown in Table 1 with different lengths can be extracted. The image blocks can be reconstructed from the compressed codes, and the decompressed image can be obtained. The detailed decoding procedures can be referred to [16].

3. Proposed Method The traditional AMBTC compression method uses the same number of bits to compress each block. However, coding in this way requires more bits than necessary for flat blocks and neglects too much image detail for complex blocks. Xiang et al.’s method improves AMBTC, resulting in better compression effects for both flat and complex blocks. However, in the processing of complex blocks, Xiang et al.’s method reconstructs the gray values of the image block by four quantized values. Although the quality of the reconstructed block is improved, it requires quantized values to be recorded and bitmaps with more bits. In addition, Xiang et al. adopt the traditional AMBTC method to compress the smooth blocks, which may increase the cost of recording bitmaps and quantized values.

In this paper, we propose a more effective solution by classifying image blocks into flat, smooth, and complex blocks based on thresholds_τ0and_τ1(_τ0≤_τ1). Let_Φi={_ai,_bi,_Bi}be the AMBTC codes of_Ii. If_bi−_ai≤_τ0,_Iiis classified as a flat block. Because pixel variations in a flat block are small, all pixels in a flat block can be simply reconstructed by their mean to a satisfactory visual quality. If_τ0<_bi−_ai<_τ1,_Iiis classified as a smooth block. For the smooth block, we use a clustering algorithm to obtain representative bitmaps, and the original bitmaps are replaced by the indices pointing to the obtained bitmap. The two quantized values are also adjusted to reduce the error caused by the bitmap replacement. If_bi−_ai≥_τ1,_Iiis classified as a complex block. We use three quantized values and a ternary map to represent the complex block to maintain better texture details. The encoding algorithms of these three types of blocks will be presented in the following sections.

3.1. Encoding of Flat Blocks

The pixel values of a flat block_Ii(i.e.,_bi−_ai≤_τ0) are relatively close, and thus the bitmap plays an insignificant role in reconstructing the image block. Therefore, we omit the recording of the quantization value in addition to the bitmap, and use an 8-bit mean value_(mi)2to represent the flat block, where:

_mi=round(_bi+_ai2)

andround(x)is the function roundingxto the nearest integer.

3.2. Encoding of Smooth Blocks

If_τ0<_bi−_ai<_τ1, the fluctuation of pixel values of block_Iiis more than that of a flat block. Therefore, we refer to_Iias a smooth block. To reduce the bitrate, a codebook consisting of thek most representative bitmaps (codewords) is found, and the bitmap of the smooth block will be replaced by an index pointing to one of the codewords in the codebook. We use the k-means algorithm [20] to obtain thekmost representative bitmaps. Let_{{as,bs,Bs}s=0Ns−1}be the set of AMBTC codes satisfying_τ0<_bi−_ai<_τ1for0≤i≤N−1, where_Nsis the number of smooth blocks. Firstly, an initial codebook_{{Cα0}α=0k−1}is constructed by randomly selectingkbitmaps from_{{Bs}s=0Ns−1}, wherekis much less than_Ns. Secondly, the bitmaps_{{Bs}s=0Ns−1}are classified intokclusters according to the similarities between_{{Bs}s=0Ns−1}and_{{Cα0}α=0k−1}. That is, if_Bshas more bits identical to_Cα0than other codewords, then_Bsis classified into groupα, where0≤α≤k−1. Thirdly,_{{Bs}s=0Ns−1}of the same group are averaged and rounded to obtain the updated codebook_{{Cα1}α=0k−1}. Repeat the classification processttimes and the final representative bitmaps_{{Cαt}α=0k−1}are obtained. Normally, settingt=6can already obtain a satisfactory result. We denote the final representative bitmaps as_{{Cα}α=0k−1}. Once the classification process is completed, the classification results_{{αs∗}s=0Ns−1}of bitmaps_{{Bs}s=0Ns−1}are also obtained. Note that the codeword with index_αs∗has the nearest distance to_Bs, that is:

_αs∗=argminα^{(∑j=0n×n−1(_Bs,j−_Cα,j)2)1/2}

where_Bs,jand_Cα,jrepresent thej-thelement of_Bsand_Cα, respectively. Instead of recording_{{Bs}s=0Ns−1}, the proposed method uses the binary representation of_{{αs∗}s=0Ns−1}as the required bitmap information. Therefore, the bits required to record the bitmap are reduced fromn×nbits to_log2(k)bits. To successfully decode the bitmap, we must have cluster centers_{{Cα}α=0k−1}and cluster indices_{{αs∗}s=0Ns−1}. Therefore,_{{Cα}α=0k−1}must be included as part of the compressed codes.

When decoding a smooth block, because we use cluster center_Cαs∗to replace the original bitmap_Bs , the quality of the reconstructed image block will be reduced. To minimize the reduced quality, a quantized value adjustment (QA) technique [18] is employed. QA is a technique originally used in a data hiding technique to reduce the distortions of the reconstructed AMBTC block when the original bitmap is replaced by secret data. Because bits in the bitmap are altered, distortions of the reconstructed block are inevitable. QA subtly adjusts the quantized values by counting the bit difference between the original bitmap and secret data. In the proposed method, the original bitmap is replaced by a cluster center, which resembles the situations in which the bitmap is replaced by secret data. Therefore, the QA technique can be applied in the proposed method. To find the minimum distortion, the QA technique adjusts_asand_bsto_a^sand_b^sby calculating:

_a^s=_{as ρ00}+_{bs ρ10ρ00}+_ρ10

and:

_b^s=_{as ρ01}+_{bs ρ11ρ01}+_ρ11

respectively, where_ρpqis the number of bits with_Bs,j=pand_Cαs∗,j=q,(p,q)∈{0,1}. For example,_ρ01indicates the number of bits with_Bs,j=0and_Cαs∗,j=1. After adjustment of quantized values, the distortion due to the bitmap replacement will be smaller than that without adjustment.

3.3. Encoding of Complex Blocks

Blocks with_bi−_ai≥_τ1are classified as complex blocks. Let_{{Ic}c=0Nc−1}be the set of_Nccomplex blocks inI. For a given complex block_Ic=_{{Ic,j}j=0n×n−1}, the proposed method uses the k-means clustering algorithm to obtain three most representative quantized values{_qc0,_qc1,_qc2}and a ternary map_Tc=_{{Tc,j}c=0n×n−1}, where_Tc,jis a ternary digit ranging from 0 to 2 used to indicate which quantized value should be used to reconstruct thej-thpixel of_Ic. Because the value of_Tc,jis equally distributed over 0 to 2, we can simply encode the ternary digits₀₃,₁₃, and₂₃by₀₂,₁₀₂, and₁₁₂, respectively. We assume the encoded result of_Tcis _^T′cof L-bit. Once the decoder has_{{ ^T′c}c=0Nc−1}and_{{qc0,qc1,qc2}c=0Nc−1}, blocks_{{Ic}c=0Nc−1}can be reconstructed.

When encoding a4×4ternary map, the average number of bits required in the proposed method is:

1×163+2×2×163=26.67 bits.

Theoretically, recording 16 ternary digits requires⌈16×_log23⌉=26bits, which is almost the same as in the proposed method. Therefore, the encoding of the ternary map used in the proposed method is effective.

3.4. Encoding Procedures

This section describes the procedures of the proposed method. To distinguish the encoding methods of three types of image blocks, an indicator is prepended to the code stream of each encoded block. The indicators₀₂,₁₀₂, and₁₁₂are used to indicate a flat, smooth, and complex block is encoded, respectively. The detailed encoding procedures are shown as follows:

Input:

Original imageI, block sizen×n, thresholds_τ0and_τ1, parameterγ, and cluster sizek.

Output:

Code streamC_Sf.

Step 1:

Partition the original imageIinto blocks_{Ii}i=0N−1of sizen×n. Encode_{Ii}i=0N−1using the AMBTC encoder and obtain codes_Φi=_{{ai,bi,Bi}i=0N−1} , as described in Section 2.1.

Step 2:

Scan codes_{{ai,bi,Bi}i=0N−1}. Let_{{Bs}s=0Ns−1}be the bitmap of smooth blocks. Clustering_{{Bs}s=0Ns−1}intokgroups using the k-means clustering algorithm, we obtainkcluster centers_{{Cα}α=0k−1}and_Nscluster indices_{{αs∗}s=0Ns−1}. Concatenate the binary representation of_{{Cα}α=0k−1}and obtain the concatenated code streamC_SA. The_Nspairs of adjusted quantized values_{{a^s,b^s}s=0Ns−1} of smooth blocks are also obtained, as described in Section 3.2. Similarly, quantized values_{{qc0,qc1,qc2}c=0Nc−1}and ternary maps_{{Tc}c=0Nc−1} of complex blocks are also obtained, as described in Section 3.3.

Step 3:

Scan codes_{{ai,bi,Bi}i=0N−1}again and perform the encoding according to the cases listed below:

Case 1:

If_bi−_ai≤_τ0, a flat block is visited and the code stream of block_IiisC_Si=₀₂||(_{mi )2}.

Case 2:

If_τ0<_bi−_ai<_τ1, a smooth block is visited. Extract_a^s,_b^s, and_αs∗from_{{a^s,b^s,αs∗}s=0Ns−1}obtained in Step 2, and block_Iiis encoded byC_Si=₁₀₂ || (_{a^s )2}||_(b^s−a^s)2||_(αs∗)2. Note that_(b^s−a^s)2 is encoded using the DES, as described in Section 2.2.

Case 3:

If_bi−_ai≥_τ1, block_Iiis a complex one. Extract_qc0,_qc1,_qc2, and_Tcfrom_{{qc0,qc1,qc2,Tc}c=0Nc−1}obtained in Step 2, and block_Iiis encoded byC_Si=₁₁₂ || (_{qc0 )2}||_(qc1−qc0)2||_(qc2−qc1)2|| _^T′c. Note that_(qc1−qc0)2and_(qc2−qc1)2 are encoded using the DES (see Section 2.2).

Step 4:

Repeat Step 3 until the code stream_{{CSi}i=0N−1}of blocks_{Ii}i=0N−1are obtained. Concatenate_{{CSi}i=0N−1}, we have the concatenated code streamC_SB.

Step 5:

ConcatenateC_SAandC_SB; we obtain the final code streamC_Sfof imageI, i.e.,C_Sf=C_SA||C_SB.

The encoding of a given image block and the number of required bits for each block types are shown in Figure 1.

We take a simple example to illustrate the encoding of smooth and complex blocks. Let_I0be a4×4 block to be encoded, as shown in Figure 2a. Suppose_τ0=4,_τ1=16,γ=64,σ=128, andk=128are used in this example. The AMBTC compressed code of_I0is{_a0,_b0,_B0}={28,40,111011101100₁₁₀₀₂}, and_B0 is depicted in Figure 2b. Because_τ0<_b0−_a0<_τ1,_I0is a smooth block. Assume_α0∗=43and_C43=1010 0110 0101 ₀₁₀₀₂ (see Figure 2c). By comparing_B0and_C43, we have_ρ00=5,_ρ01=1,_ρ10=4, and_ρ11=6. Using Equations (7) and (8), we have_a^0=33and_b^0=38. Because_b^0−_a^0=5<γ, we haveY=0. Because_(a^0)2=_001000012,_(b^0−a^0)2=0||_0001012, and_(α0∗)2=_01010112, the code stream of_I0should beC_S0=10 || 00100001 || 0||000101 || _01010112.

Figure 2d shows another block_I1to be encoded. For this block, quantized values_a1=25and_b1=103of the AMBTC code are calculated. Because_b1−_a1≥_τ1,_I1is regarded as a complex block. Suppose after applying the k-means clustering algorithm to_I1, we obtain three quantized values{_q10,_q11,_q12}={19, 85, 133}and the ternary cluster indices of pixels_T1={1111 2121 0210 0000} , as shown in Figure 2e. The difference between the first two quantized values is_q11−_q10 = 66>γ. Therefore, indicator_Y1,0=1should be placed in front of the_log2(σ)=7-bitbinary representation of 66 (i.e.,1||_10000102). Similarly, because_q12−_q11 = 48<γ, indicator_Y1,1=0should be placed in front of the_log2(γ)=6-bitbinary representation of 48 (i.e.,_0||1100002). Finally, the ternary cluster indices_T1are encoded by10101010 11101110 011100 ₀₀₀₀₂ , which is illustrated in Figure 2f. Therefore, according to Step 3 of Case 3 in Section 3.4, the code stream of block_I1should beC_S1=11 || 00010011 || 1||1000010 || 0||110000 || 10101010 11101110 _{011100 00002}.

3.5. Decoding Procedures In decoding, data bits are sequentially read and decoded, and image blocks are reconstructed by decoding the read data bits. The detailed steps of decoding are listed as follows:

Input:

Code streamC_Sf, block sizen×n, parameterγ,σ, and cluster sizek.

Output:

Decompressed image^I′=_{{Ii′}i=0N−1}.

Step 1:

ExtractC_SAfromC_Sfand reconstructkcluster centers_{{Cα}α=0k−1}.

Step 2:

Extract one bitbfromC_Sf. According to the extracted bit, one of the following decoding cases is then performed:

Case 1:

Ifb=₀₂, the block to be reconstructed is a flat block. All the pixel values of block_Ii′are the decimal value of the next 8 bits extracted fromC_Sf.

Case 2:

Ifb=₁₂and the next extracted bit is₀₂, the block to be reconstructed is a smooth block. Extract the next 8 bits and convert them to a decimal value to obtain the quantized value_a^s. Read the next bit fromC_Sf. If the read bit is₀₂,_b^sis reconstructed by the decimal value of the next_log2(γ)bits plus_a^s. Otherwise,_b^sis reconstructed by the decimal value of next⌈_log2(σ)⌉bits plus_a^s. The clustering index_αs∗is the decimal value of nextkbits, and the bitmap_Cαs∗can be obtained from_{{Cα}α=0k−1}. Using the AMBTC decoder to decode{_a^s,_b^s,_Cαs∗}, the image block can be reconstructed.

Case 3:

Ifb=₁₂and the next extracted bits is₁₂, the block to be reconstructed is a complex block. Extract the next 8 bits and convert them to a decimal value to obtain the quantized value_qc0. Read the next bit fromC_Sf. If the read bit is₀₂,_qc1is reconstructed by the_qc0plus the decimal value of the next_log2(γ)bits; otherwise,_qc1is reconstructed by the decimal value of the next⌈_log2(σ)⌉bits plus_qc0. Using a similar manner,_qc2is reconstructed. To reconstruct the ternary map_{{Tc,j}j=0n×n−1}, we start fromj=0toj=n×n−1and repeat the following process: Read a bit_b0fromC_Sf. If_b0=₀₂, we have_Tc,j=0. Otherwise, read the next bit_b1fromC_Sf. If_{b0 b1}=₁₀₂,_Tc,j=1. If_{b0 b1}=₁₁₂,_Ti,j=2. Once we have{_qc0,_qc1,_qc2}and_{{Tc,j}j=0n×n−1}, the j-th pixel of the image block is reconstructed by_qc0,_qc1, or_qc2if_Tc,j=0, 1, or 2, respectively.

Step 3:

Repeat Step 2 until all image blocks are reconstructed, and the final decompressed image^I′is obtained.

We continue the example given in Section 3.4 to illustrate the decoding process. The detailed process and the decoded result are depicted in Figure 3. To decode the code streamC_S0=10 || 00100001 || 0 000101 || _01010112, because the first bit is₁₂and the second bit is₀₂, the to-be-reconstructed block is a smooth block. Extract the next 8 bits fromC_S0and convert them into decimal representation; we obtain_a^0=33. The next extracted bit is 0. Therefore, the difference_d0=5is the decimal value of the next_log2(γ)bits, and we have_b^0=_a^0+5=38. Finally, extract_log2(k)bits and convert them to a decimal value; we have_αs∗=43and the bitmap_C43is obtained. The image block can then be constructed by decoding{_a^0,_b^0,_C43}using the AMBTC decompression technique.

To decodeC_S1=11 || 00010011 || 1 1000010 || 0 110000 || 10101010 11101110 _{011100 00002}, because the first two extracted bits are₁₁₂, the block to be decompressed is a complex block. Extract 8 bits and_q10=19is the decimal value of these 8 bits. The next bit is₁₂; therefore,_d11=66is the decimal value of the next⌈_log2(σ)⌉=7bits and_q11=19+66=85can be obtained. Similarly, the next extracted bit is₀₂; therefore,_d12=48is obtained by converting the next_log2(γ)=6bits to their decimal value, and_q12=85+48=133can be obtained. Finally, we have to reconstruct_{T1,j}j=015from the remaining bits. Because the next extracted bit is₁₂, we extract one more bit, which is₀₂. Therefore,_T1,0=1is obtained. The remaining 15 ternary digits_{T1,j}j=115can be decoded in the similar manner. Once we have{_q10,_q11,_q12}and_{T1,j}j=015, block_I1′ can be reconstructed. Figure 3b illustrates the decoding process ofC_S1.

4. Experimental Results

In this section, we conduct several experiments to show the effectiveness and applicability of the proposed scheme. We take eight grayscale images of size512×512 , namely, Lena, Jet, Baboon, Tiffany, Boat, Stream, Peppers and House, as the test images, as shown in Figure 4. These images can be obtained from the USC-SIPI image database [21]. We use the peak signal-to-noise ratio (PSNR) and bitrate to measure the performance. The PSNR is calculated by:

PSNR=10_log10²⁵⁵²1n×n×N∑_{i=0n×n×N−1} ^{(_xi− _x′i)2},

where_xiand _^x′irepresent the pixel values of the original and decompressed images, respectively. The bitrate metric is measured by the number of bits required to record each pixel (i.e., bit per pixel, bpp).

In all of the experiments, we set_τ0=4because the flat blocks under this setting show no apparent block boundary artifacts.

4.1. The Performance of the Proposed Method

Because the number of cluster centers k and threshold greatly affect the coding efficiency in the application of the quantized value adjustment (QA) technique [18], we evaluate how the QA technique and these parameters influence the bitrate and image quality in this section.

4.1.1. Coding Efficiency Comparisons

In the coding of smooth blocks, the original bitmaps of smooth blocks are used to obtain the cluster centers, and the quantized values are adjusted using the QA technique to lower the distortions. Table 2 and Table 3 show how the QA technique improves the image quality when the block size is set to4×4and8×8, respectively. In this experiment,_τ1=16andγ=32are set. As seen from the tables, the QA technique effectively enhances the quality of reconstructed images for everyk . For example, in Table 2 andk=128, the averaged quality of the reconstructed images with and without the QA technique is 34.63 and 34.52 dB, respectively. The averaged quality has improved by34.63−34.52=0.11 dB. Similarly, in Table 3 whenk=128, the PSNR improvement is31.86−31.76=0.10dB. Therefore, the QA technique indeed reduces the distortion caused by replacing the original bitmap with a cluster center.

Table 2 and Table 3 also reveal that the increase in cluster sizek also enhances the image quality. For example, in Table 3 whenk=128, the averaged PSNR of eight test images is 31.86 dB. Whenk=256and 512, the PSNR increases31.91−31.86=0.05dB and32.00−31.86=0.14dB, respectively. The reason is that a larger cluster size provides a greater chance to reduce the difference between the cluster centers and original bitmaps.

To evaluate how threshold_τ1affects the performance of the QA, we plot the gain of PSNR when using the QA for various_τ1withk=128 and 512. The results are shown in Figure 5. Note that, in this experiment, a block size of4×4is set.

Figure 5a,b shows that the gain in PSNR increases as_τ1increases, and this is mainly because the number of smooth blocks also increases as_τ1increases. Because more blocks are classified as smooth for larger_τ1, more blocks will be processed using the QA technique. As a result, the gain in PSNR is higher when_τ1is larger. It also can be observed that for each test image, the gain in PSNR is larger whenk=128than that whenk=512. The reason is that a smallerkimplies larger differences between the original bitmaps and cluster centers. Because the QA technique is capable of reducing the distortion caused by the differences, a larger PSNR improvement can be achieved for smallerk.

It is interesting to note that the gain in PSNR of the Stream and Baboon images increases more than that of other test images when varying_τ1=10to_τ1=50for bothk=128andk=512. Because these two images are more complex than the others, their bitmaps of smooth blocks are expected to be more different from the selected cluster centers used to replace the bitmaps. As previously mentioned, the QA technique is effective in reducing the distortion caused by the differences, and the bitmaps of Stream and Baboon images are more different from the cluster centers than the other images. Therefore, the improvement in PSNR after applying the QA technique is more significant than for the others.

4.1.2. Performance Comparison of Various_τ1

The parameter_τ1controls the number of smooth and complex blocks. The number of complex blocks decreases as_τ1increases. To see the distribution of flat, smooth, and complex blocks of a test image, we take the Lena image as an example to illustrate their distribution by varying_τ1 . Figure 6a–d shows the distributions of blocks when_τ1=8, 16, 32, and 64 are set. The block sizes in these figures are8×8and_τ0=4. In this figure, the blue squares, red dots, and black cross marks represent flat, smooth, and complex blocks, respectively.

Because the same_τ0 is applied, it can be seen that the number of blue squares (flat blocks) is the same in Figure 6a–d. However, as_τ1increases, the red dots increase and black cross marks decrease. The reason is that an increase in_τ1leads more blocks to be categorized as smooth. It can also be inferred that a better image quality can be achieved at a smaller_τ1but the bitrate will be higher because more blocks are deemed to be complex. Note that in the proposed method, more bits are required to represent a complex block than a smooth block.

Table 4 and Table 5 show the PSNR and bitrate for all of the test images under various_τ1with block size4×4and8×8, respectively. In this experiment,_τ0=4,γ=64, andk=256are set. We also list the PSNR and bitrate of the standard AMBTC method as a comparison. Note that the bitrates of the AMBTC with block size4×4and8×8are 2.0 and 1.25 bpp, respectively.

As seen in Table 4 and Table 5, the PSNR of block size8×8is lower than that of block size4×4. For example, when_τ1=16, the PSNR of the Lena image of block sizes4×4and8×8are 34.73 and 31.91 dB, respectively. However, the former requires1.68−1.02=0.66more bits per pixel than the latter. In addition, the experiments also reveal the fact that a large_τ1effectively reduces the bitrate at the expense of image quality. On the contrary, a small_τ1provides better image quality, but requires more bitrate. This result is expected because a small_τ1increases the number of complex blocks and, therefore, the bitrate, however, the image quality also increases.

Figure 7 shows the bitrate–PSNR curves of each of the eight test images by varying the threshold_τ1from 8 to 64. The figure shows that for all of the test images, the PSNR increases as the bitrate increases. Moreover, the figure also reveals that smooth images, such as Tiffany or Jet, have a better compression efficiency than those of complex images, such as Stream or Baboon. The reason is that a smooth block not only requires less bits to record its compressed code but also provides better reconstructed quality. Because the smooth images naturally possess more smooth blocks than complex blocks, their bitrate–PSNR curves are higher than those of complex ones.

It also can be seen from Figure 7 that the PSNR and bitrate vary as the threshold_τ1changes. A larger_τ1gives a lower bitrate with lower PSNR. In contrast, a smaller_τ1offers a higher image quality, but the bitrate is also higher. Therefore, the selection of threshold_τ1depends on real applications. For example, if an application requires higher image quantity, a smaller_τ1is required.

It is worth noting that, for most of the test images, the proposed method provides better performance than AMBTC, particularly for smooth images. For example, Lena, Jet, Tiffany, Boat, and Peppers are considered to be smooth. For these smooth images, regardless of the value of_τ1, the PSNR is always higher and the bitrate is always lower than those of the AMBTC method. In contrast, for the complex images such as Baboon or Stream, few blocks are classified as flat, which require only 8 bits to record them. Therefore, the reduction in bitrate is limited. Nevertheless, the proposed method either provides a better image quality or lower bitrate than those of AMBTC.

4.2. Comparisons with Xiang et al.’s Work

Xiang et al.’s method [16] also improves the AMBTC method by dynamically splitting images into multiple groups and achieves a good performance. In this section, we compare the proposed method with that of Xiang et al. in terms of PSNR and bitrate. To make a fair comparison, threshold_τ0=4andγ=64are set in both methods. The proposed method uses_τ1to control the number of smooth and complex blocks, whereas Xiang et al.’s method uses_dminto control the number of pixel groups. We select_τ1=8, 16, and 24 in the proposed method and compare the results with those of Xiang et al. by setting_dmin=6, 7, and 8 for block size4×4 . The results are shown in Table 6. Table 7 shows the same experimental results, except block size is8×8and_dmin=28, 32, and 36. The settings of_dminin Xiang et al.’s method ensure that best performance can be achieved.

Table 6 shows that in Xiang et al.’s method, as_dminincreases, the image quality and bitrate decrease. The reason is that a large_dminprevents more blocks from being split, leading to a decrease in bitrate and PSNR. Note that for most of the test images, the proposed method performs better than that of Xiang et al. We take the Lena image as an example: when_τ1=16and_dmin=7are set, the PSNR of the proposed and Xiang et al.’s methods are 36.05 dB with 1.68 bpp and 35.12 dB with 2.20 bpp, respectively. The PSNR of the proposed method is36.05−35.12=0.93dB higher and the bitrate is2.20−1.68=0.52bpp lower than that of Xiang et al.’s method. Comparisons with other images and another set of parameters also reveal similar results, with the exception of the Baboon image. When_τ1=8and_dmin=6are set, the PSNR of Xiang et al.’s method is31.40−31.28=0.12dB higher than the proposed method. The reason is that under these settings, more blocks are divided into four groups and thus a better image quality is achieved. However, their method requires3.57−3.12=0.45bpp more than the proposed method.

In the performance comparisons with block size8×8 , the proposed method shows better results for all test images. For example, as shown in Table 7 when_τ1=14and_dmin=28, the PSNR of the Baboon image of the proposed method is 28.84 dB at 1.88 bpp. The PSNR is28.84−28.14=0.70dB higher and the bitrate is2.23−1.88=0.35bpp lower than those of Xiang et al.’s method.

Figure 8a–f shows the visual quality comparisons of the AMBTC, Xiang et al.’s, and the proposed methods. As seen from Figure 8a when the block size is4×4 and the AMBTC method is applied, apparent distortions can be seen in the image edges, and noticeable boundary artifacts are observed (see Figure 8a). Note that the PSNR of the AMBTC is 33.27 dB with 2.0 bpp. Xiang et al. improve the AMBTC method by adding more details to complex blocks. As a result, the PSNR (36.31 dB) is significantly higher and blocks at the edges look more natural than those of AMBTC (Figure 8c,_dmin=6). However, their method requires2.35−2=0.35 bpp more to achieve this effect. In contrast, the visual quality of the proposed method (Figure 8e,_τ1=8) is comparable with that of Xiang et al.’s method, but the bitrate is2.35−2.05=0.30bpp lower with a slightly higher PSNR.

When the block size is8×8 , the distortion of AMBTC is more apparent (Figure 8b) than that of4×4 , but the bitrate reduces from 2.0 to 1.25 bpp. The visual quality of Xiang et al.’s method (Figure 8d,_dmin=28) is significantly better than that of AMBTC, and has no noticeable block boundary artifacts. However, their method requires1.64−1.25=0.39bpp more to improve the image quality. In addition, some edges in Lena’s face, eyes, and shoulder exhibit apparent distortions because the pixel splitting operation may not be triggered due to the setting of_dmin . In contrast, the edges of the proposed method exhibit no apparent distortion (see Figure 8f,_τ1=14). Moreover, the bitrate required in the proposed method is even lower than that of AMBTC by1.25−1.09=0.16bpp.

5. Conclusions In this paper, we propose a hybrid encoding scheme for AMBTC compressed images using a ternary representation technique. Considering that the number of quantized values greatly affects the quality of the reconstructed image, the proposed method classifies image blocks into flat, smooth, and complex. These three types of blocks are encoded by using one, two, or three quantized values. Flat blocks require no bitmap, whereas smooth and complex blocks require binary and ternary maps, respectively, to record the quantized values to be used to reconstruct the corresponding pixels. A sophisticated design indicator is prepended before the code stream of a block to signify the block type. The proposed method achieves a better image quality than that of prior works with a smaller bitrate. The effectiveness of the proposed method is observed from the experimental results. Note that although the k-means algorithm used in the proposed method may require slightly higher computational cost than that of the discrete cosine transform (DCT) based methods, it is only applied to smooth blocks in the encoding stage to obtain the representative bitmaps rather than the whole image. Furthermore, the k-means algorithm does not need to be applied again during decoding. Therefore, the overall computational cost of the proposed method is smaller than that of DCT-based compression methods.

Figure 1. Illustration of image block encoding.

Figure 2. An example of image encoding.

Figure 2. An example of image encoding.

Figure 3. Illustration of the decoding procedures of image blocks.

Figure 4. Eight test images.

Figure 5. The gain of PSNR when the quantized value adjustment (QA) technique is applied.

Figure 6. Distribution of flat, smooth and complex blocks.

Figure 7. Bitrate-PSNR curves of each of the eight test images.

Figure 8. Visual quality comparisons of the proposed method and that of Xiang et al.

Number of Groups	Compressed Code of Ii	Number of Bits
1	{002,(mi)2}	2+8
2	{012,(ai )2, (bi−ai)2, Bi}	2+8+Ri+n×n
3	{102,(ai0 )2, (bi0−ai0)2, (bi−bi0)2, Bi, Ji, Bi0}or	2+8+2Ri+n×n+1+Pi
	{102,(ai )2 ,(ai1−ai)2 ,(bi1−ai1)2 ,Bi, Ji, Bi1}
4	{112,(ai0)2, (bi0−ai0)2, (ai1−bi0)2, (bi1−ai1)2, Bi, Bi0,Bi1}	2+8+3Ri+2×n×n

Images	k=128			k=256			k=512
	Bitrate	w/o QA	w/QA	Bitrate	w/o QA	w/QA	Bitrate	w/o QA	w/QA
Lena	1.60	35.83	35.96	1.63	35.95	36.05	1.68	36.07	36.15
Jet	1.52	35.85	35.91	1.55	35.92	35.97	1.58	35.99	36.03
Baboon	2.73	30.80	30.85	2.76	30.85	30.89	2.79	30.89	30.93
Tiffany	1.47	37.18	37.38	1.50	37.39	37.55	1.55	37.57	37.69
Boat	2.00	33.97	34.13	2.05	34.15	34.27	2.10	34.27	34.36
Stream	2.68	32.43	32.49	2.70	32.49	32.53	2.73	32.55	32.57
Peppers	1.68	35.34	35.57	1.73	35.52	35.71	1.78	35.71	35.85
House	1.95	34.73	34.78	1.97	34.79	34.82	2.01	34.85	34.88
Average	1.95	34.52	34.63	1.99	34.63	34.72	2.03	34.74	34.81

Images	k=128			k=256			k=512
	Bitrate	w/o QA	w/QA	Bitrate	w/o QA	w/QA	Bitrate	w/o QA	w/QA
Lena	0.97	32.96	33.07	1.01	33.03	33.12	1.08	33.13	33.21
Jet	1.00	32.77	32.82	1.04	32.83	32.86	1.11	32.90	32.92
Baboon	1.79	28.67	28.72	1.83	28.72	28.75	1.89	28.80	28.82
Tiffany	0.89	34.63	34.81	0.93	34.79	34.93	1.00	34.99	35.09
Boat	1.30	31.17	31.32	1.34	31.24	31.37	1.41	31.36	31.46
Stream	1.89	29.79	29.82	1.92	29.84	29.86	1.99	29.93	29.93
Peppers	1.01	32.59	32.74	1.05	32.66	32.80	1.12	32.79	32.90
House	1.36	31.52	31.57	1.39	31.57	31.61	1.46	31.65	31.67
Average	1.28	31.76	31.86	1.31	31.83	31.91	1.38	31.94	32.00

Images	AMBTC bpp = 2.0	PSNR (dB)			Bitrate (bpp)
		τ1=16	τ1=32	τ1=64	τ1=16	τ1=32	τ1=64
Lena	33.24	36.05	34.62	32.85	1.68	1.46	1.33
Jet	31.97	35.98	34.79	32.50	1.56	1.39	1.24
Baboon	26.98	30.89	29.55	25.74	2.75	2.34	1.75
Tiffany	35.77	37.55	35.72	34.26	1.54	1.34	1.25
Boat	31.16	34.27	32.87	30.81	2.09	1.77	1.57
Stream	28.59	32.53	30.20	27.27	2.73	2.11	1.66
Peppers	33.42	35.71	34.49	32.80	1.77	1.59	1.49
House	30.89	34.82	32.00	30.42	2.00	1.65	1.38
Average	31.50	34.73	33.03	30.83	2.02	1.71	1.46

Images	AMBTC bpp = 1.25	PSNR (dB)			Bitrate (bpp)
		τ1=16	τ1=32	τ1=64	τ1=16	τ1=32	τ1=64
Lena	29.93	33.12	31.80	29.24	1.02	0.76	0.51
Jet	28.84	32.86	31.72	29.40	1.04	0.82	0.62
Baboon	25.18	28.75	27.66	23.06	1.81	1.43	0.71
Tiffany	32.55	34.93	32.76	30.74	0.94	0.62	0.47
Boat	28.07	31.37	29.78	27.44	1.35	0.90	0.61
Stream	26.10	29.85	27.99	24.18	1.92	1.34	0.63
Peppers	29.66	32.80	31.49	29.43	1.06	0.78	0.58
House	27.68	31.60	30.20	26.92	1.39	1.02	0.62
Average	28.50	31.91	30.43	27.55	1.32	0.96	0.59

Images	Metrics	Proposedτ1=8	[16]dmin=6	Proposedτ1=16	[16]dmin=7	Proposedτ1=24	[16]dmin=8
Lena	PSNR	36.98	36.31	36.05	35.12	35.31	33.91
	Bitrate	2.05	2.35	1.68	2.20	1.54	1.95
Jet	PSNR	36.53	34.96	35.97	33.74	35.36	32.69
	Bitrate	1.80	1.98	1.57	1.87	1.45	1.71
Baboon	PSNR	31.28	31.40	30.89	29.59	30.32	28.11
	Bitrate	3.12	3.57	2.75	3.24	2.52	2.75
Tiffany	PSNR	38.98	38.42	37.55	37.29	36.46	36.41
	Bitrate	1.92	2.12	1.54	1.99	1.40	1.81
Boat	PSNR	35.37	34.81	34.27	33.43	33.44	32.11
	Bitrate	2.65	3.02	2.09	2.79	1.87	2.43
Stream	PSNR	32.98	32.69	32.53	31.14	31.46	29.69
	Bitrate	3.07	3.45	2.73	3.18	2.39	2.75
Peppers	PSNR	37.11	36.38	35.71	35.23	35.03	34.16
	Bitrate	2.28	2.64	1.77	2.45	1.65	2.20
House	PSNR	35.32	35.06	34.82	33.52	33.99	31.86
	Bitrate	2.28	2.52	2.00	2.35	1.80	2.05

Images	Metrics	Proposedτ1=14	[16]dmin=28	Proposedτ1=22	[16]dmin=32	Proposedτ1=30	[16]dmin=36
Lena	PSNR	33.32	31.88	32.66	30.85	32.00	30.27
	Bitrate	1.09	1.64	0.90	1.45	0.79	1.27
Jet	PSNR	32.99	30.37	32.44	29.66	31.87	29.23
	Bitrate	1.09	1.36	0.94	1.24	0.84	1.13
Baboon	PSNR	28.84	28.14	28.43	26.44	27.84	25.50
	Bitrate	1.88	2.23	1.65	1.88	1.47	1.50
Tiffany	PSNR	35.36	34.44	33.97	33.48	32.95	32.89
	Bitrate	1.04	1.51	0.77	1.33	0.64	1.16
Boat	PSNR	31.62	30.12	30.68	29.11	29.95	28.54
	Bitrate	1.46	1.99	1.11	1.75	0.93	1.50
Stream	PSNR	29.94	28.62	29.39	27.30	28.31	26.54
	Bitrate	1.99	2.20	1.71	1.90	1.41	1.58
Peppers	PSNR	33.01	31.35	32.27	30.52	31.68	30.03
	Bitrate	1.14	1.86	0.91	1.62	0.81	1.40
House	PSNR	31.71	30.20	31.22	28.79	30.45	28.09
	Bitrate	1.45	1.70	1.24	1.47	1.06	1.26

Author Contributions

W.H., J.W., and K.S.C. contributed to the conceptualization, methodology, and writing of this paper. J.Y. and T.-S.C conceived the simulation setup, formal analysis and conducted the in-vestigation. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Data is contained within the article.

Conflicts of Interest

The authors declare no conflict of interest.

References

1. Liu, J.; Tian, Y.-G.; Han, T.; Yang, C.-F.; Liu, W.-B. LSB steganographic payload location for JPEG-decompressed images. Digit. Signal Process. 2015, 38, 66-76.

2. Liu, J.; Tian, Y.; Han, T.; Wang, J.; Luo, X. Stego key searching for LSB steganography on JPEG decompressed image. Sci. China Inf. Sci. 2016, 59, 1-15.

3. Qin, C.; Chang, C.-C.; Chiu, Y.-P. A Novel Joint Data-Hiding and Compression Scheme Based on SMVQ and Image Inpainting. IEEE Trans. Image Process. 2014, 23, 969-978.

4. Qin, C.; Hu, Y.-C. Reversible data hiding in VQ index table with lossless coding and adaptive switching mechanism. Signal Process. 2016, 129, 48-55.

5. Tsou, C.-C.; Hu, Y.-C.; Chang, C.-C. Efficient optimal pixel grouping schemes for AMBTC. Imaging Sci. J. 2008, 56, 217-231.

6. Hu, Y.C.; Su, B.H.; Tsai, P.Y. Color image coding scheme using absolute moment block and prediction technique. Imaging Sci. J. 2008, 56, 254-270.

7. Delp, E.J.; Mitchell, O.R. Image coding using block truncation coding. IEEE Trans. Commun. 1979, 27, 1335-1342.

8. Lema, M.; Mitchell, O. Absolute Moment Block Truncation Coding and Its Application to Color Images. IEEE Trans. Commun. 1984, 32, 1148-1157.

9. Kumaravadivelan, A.; Nagaraja, P.; Sudhanesh, R. Video compression technique through block truncation coding. Int. J. Res. Anal. Rev. 2019, 6, 236-242.

10. Hemida, O.; He, H. A self-recovery watermarking scheme based on block truncation coding and quantum chaos map. Multimed. Tools Appl. 2020, 79, 18695-18725.

11. Qin, C.; Ji, P.; Zhang, X.; Dong, J.; Wang, J. Fragile image watermarking with pixel-wise recovery based on overlapping embedding strategy. Signal Process. 2017, 138, 280-293.

12. Qin, C.; Ji, P.; Chang, C.-C.; Dong, J.; Sun, X. Non-uniform Watermark Sharing Based on Optimal Iterative BTC for Image Tampering Recovery. IEEE MultiMed. 2018, 25, 36-48.

13. Ma, Y.Y.; Luo, X.Y.; Li, X.L.; Bao, Z.; Zhang, Y. Selection of rich model steganalysis features based on decision rough set α-positive region reduction. IEEE Trans. Circuits Syst. Video Technol. 2019, 29, 336-350.

14. Zhang, Y.; Qin, C.; Zhang, W.M.; Liu, F.L.; Luo, X.Y. On the fault-tolerant performance for a class of robust image ste-ganography. Signal Process. 2018, 146, 99-111.

15. Hu, Y.-C. Low-complexity and low-bit-rate image compression scheme based on absolute moment block truncation coding. Opt. Eng. 2003, 42, 1964-1975.

16. Xiang, Z.; Hu, Y.-C.; Yao, H.; Qin, C. Adaptive and dynamic multi-grouping scheme for absolute moment block truncation coding. Multimed. Tools Appl. 2018, 78, 7895-7909.

17. Chen, W.-L.; Hu, Y.-C.; Liu, K.-Y.; Lo, C.-C.; Wen, C.-H. Variable-Rate Quadtree-segmented Block Truncation Coding for Color Image Compression. Int. J. Signal Process. Image Process. Pattern Recognit. 2014, 7, 65-76.

18. Hong, W. Efficient Data Hiding Based on Block Truncation Coding Using Pixel Pair Matching Technique. Symmetry 2018, 10, 36.

19. Mathews, J.; Nair, M.S. Adaptive block truncation coding technique using edge-based quantization approach. Comput. Electr. Eng. 2015, 43, 169-179.

20. Hartigan, J.A.; Wong, M.A. A K-means clustering algorithm. Appl. Stat. 1979, 28, 100-108.

21. The USC-SIPI Image Database. Available online: http://sipi.usc.edu/database/ (accessed on 1 November 2020).