Multiple description (MD) coding is an attractive framework for robust information transmission over non-prioritized and unpredictable networks. In this paper, a novel MD image coding scheme is proposed based on convolutional neural networks (CNNs), which aims to improve the reconstructed quality of side and central decoders. For this purpose initially, a given image is encoded into two independent descriptions by sub-sampling. Such a design can make the proposed method compatible with the existing image coding standards. At the decoder, in order to achieve high-quality of side and central image reconstruction, three CNNs, including two side decoder sub-networks and one central decoder sub-network, are adopted into an end-to-end reconstruction framework. Experimental results show the improvement achieved by the proposed scheme in terms of both peak signal-to-noise ratio values and subjective quality. The proposed method demonstrates better rate central and side distortion performance.

This paper proposes a lossy image coding method for still images. In this method, recursive and non-recursive type intra prediction techniques are adaptively selected on a block-by-block basis. The recursive-type intra prediction technique applies a linear predictor to each pel within a prediction block in a recursive manner, and thus typically produces smooth image values. In this paper, the non-recursive type intra prediction technique is extended from the angular prediction technique adopted in the H.265/HEVC video coding standard to enable interpolative prediction to the maximum possible extent. The experimental results indicate that the proposed method achieves better coding performance than the conventional method that only uses the recursive-type prediction technique.

In this paper, we propose a 2-layer lossless coding method for high dynamic range (HDR) images based on range compression and adaptive inverse tone-mapping. Recently, HDR images, which have a wider range of luminance than conventional low dynamic range (LDR) ones, have been frequently used in various fields. Since commonly used devices cannot yet display HDR images, 2-layer coding methods that decode not only HDR images but also their LDR versions have been proposed. We have previously proposed a state-of-the-art 2-layer lossless coding method for HDR images that unfortunately has huge HDR file size. Hence, we introduce two ideas to reduce the HDR file size to less than that of the previous method. The proposed method achieves high compression ratio and experiments show that it outperforms the previous method and other conventional methods.

Block compressed sensing (CS) with optimal permutation is a promising method to improve sampling efficiency in CS-based image compression. However, the existing optimal permutation scheme brings a large amount of extra data to encode the permutation information because it needs to know the permutation information to accomplish signal reconstruction. When the extra data is taken into consideration, the improvement in sampling efficiency of this method is limited. In order to solve this problem, a new optimal permutation strategy for block CS (BCS) is proposed. Based on the proposed permutation strategy, an improved optimal permutation based BCS method called BCS-NOP (BCS with new optimal permutation) is proposed in this paper. Simulation results show that the proposed approach reduces the amount of extra data to encode the permutation information significantly and thereby improves the sampling efficiency compared with the existing optimal permutation based BCS approach.

We previously proposed a lossless image coding scheme using example-based probability modeling, wherein the probability density function of image signals was dynamically modeled pel-by-pel. To appropriately estimate the peak positions of the probability model, several examples, i.e., sets of pels whose neighborhoods are similar to the local texture of the target pel to be encoded, were collected from the already encoded causal area via template matching. This scheme primarily makes use of non-local information in image signals. In this study, we introduce a prediction technique into the probability modeling to offer a better trade-off between the local and non-local information in the image signals.

We present a lifting-based lapped transform (L-LT) and a reversible symmetric extension (RSE) in the boundary processing for more effective lossy-to-lossless image coding of data with various qualities from only one piece of lossless compressed data. The proposed dual-DCT-lifting-based LT (D2L-LT) parallel processes two identical LTs and consists of 1-D and 2-D DCT-liftings which allow the direct use of a DCT matrix in each lifting coefficient. Since the DCT-lifting can utilize any existing DCT software or hardware, it has great potential for elegant implementations that are dependent on the architecture and DCT algorithm used. In addition, we present an improved RSE (IRSE) that works by recalculating the boundary processing and solves the boundary problem that the DCT-lifting-based L-LT (DL-LT) has. We show that D2L-LT with IRSE mostly outperforms conventional L-LTs in lossy-to-lossless image coding.

Ultra high definition (UHD) imaging systems have attracted much attention as a next generation television (TV) broadcasting service and video streaming service. However, the state of the art video coding standards including H.265/HEVC has not enough compression rate for streaming, broadcasting and storing UHD. Existing coding standard such as H.265/HEVC normaly use RGB-YCbCr color transform before compressing RGB color image since that procedure can decorrelate color components well. However, there is room for improvement on the coding efficiency for color image based on an observation that the luminance and chrominance components changes in same locations. This observation inspired us to propose a new post-processing method for compressed images by using weighted least square (WLS) filter with coded luminance component as a guide image, for refining the edges of chrominance components. Since the computational cost of WLS tends to superlinearly increase with increasing image size, it is difficult to apply it to UHD images. To overcome this problem, we propose slightly overlapped block partitioning and a new variant of WLS (constrained WLS, CWLS). Experimental results of objective quality comparison and subjective assessment test using 4K images show that our proposed method can outperform the conventional method and reduce the bit amount for chrominance component drastically with preserving the subjective quality.

This letter introduces a new reference frame to improve the performance of motion estimation and compensation in video coding, based on a video stabilization technique. The proposed method synthesizes the new reference frame from the previous frame in a way that the new reference and current frames have the same camera orientations. The overhead data for each frame to transmit from an encoder to a decoder is only three rotational angles along the x, y, and z axes. Since the new reference and current frames have the same camera orientations, the proposed method significantly improves the performance of motion estimation and compensation for video sequences having dynamic camera motion by up to 0.98 dB with negligible overhead data.

Decimation and interpolation methods are utilized in image coding for low bit rate image coding. However, the decimation filter (prefilter) and the interpolation filter (postfilter) are irreversible with each other since the prefilter is a wide matrix (a matrix whose number of columns are larger than that of rows) and the postfilter is a tall one (a matrix whose number of rows are larger than that of columns). There will be some distortions in the reconstructed image even without any compression. The method of interpolation-dependent image downsampling (IDID) was used to tackle the problem of producing optimized downsampling images, which led to the optimized prefilter of a given postfilter. We propose integrating the IDID with time-domain lapped transforms (TDLTs) to improve image coding performance.

This paper presents an M-channel (M=2n (n ∈ N)) integer discrete cosine transforms (IntDCTs) based on fast Hartley transform (FHT) for lossy-to-lossless image coding which has image quality scalability from lossy data to lossless data. Many IntDCTs with lifting structures have already been presented to achieve lossy-to-lossless image coding. Recently, an IntDCT based on direct-lifting of DCT/IDCT, which means direct use of DCT and inverse DCT (IDCT) to lifting blocks, has been proposed. Although the IntDCT shows more efficient coding performance than any conventional IntDCT, it entails many computational costs due to an extra information that is a key point to realize its direct-lifting structure. On the other hand, the almost conventional IntDCTs without an extra information cannot be easily expanded to a larger size than the standard size M=8, or the conventional IntDCT should be improved for efficient coding performance even if it realizes an arbitrary size. The proposed IntDCT does not need any extra information, can be applied to size M=2n for arbitrary n, and shows better coding performance than the conventional IntDCTs without any extra information by applying the direct-lifting to the pre- and post-processing block of DCT. Moreover, the proposed IntDCT is implemented with a half of the computational cost of the IntDCT based on direct-lifting of DCT/IDCT even though it shows the best coding performance.

In this paper, we present a perceptual distortion measure (PDM) for polygon-based shape coding. We model the PDM as the salience of relevance triangle, and express the PDM by using three properties derived from the salience of visual part. Performance analysis and experimental results show that our proposal can improve the quality of the shape reconstruction when the object contour has sharp protrusions.

In this paper, we present a method to combine lapped transforms with various downsampling factors. The factor is changed depending on a local feature of a given signal, and it can be realized by using time-domain lapped transforms. In image coding application, our method maintains good image coding performance for a wide range of bitrates and fills the gap between undersampled and critically-/oversampled systems.

This paper presents a memory efficient version of set partitioning in hierarchical tree (SPIHT). The proposed coder called Memory Efficient SPIHT (MESH) uses a single re-usable list instead of three continuously growing linked lists as in conventional SPIHT. The list is re-initialized at the beginning of each bit-plane (coding pass) and is exhausted within the same bit-plane. Another feature of the proposed coder is that it uses a single pass for each bit-plane by merging the sorting and refinement passes of the conventional version of SPIHT. The performance of the proposed coder is measured in terms of coding efficiency, and the worst case dynamic memory requirements due to the list entries in each bit-plane. Performance comparison with SPIHT shows that the proposed coder reduces the dynamic memory requirement by about 50–70% compared to the SPIHT while retaining its coding efficiency.

A simple image compression scheme is presented for various types of images, which include color/grayscale images, color-quantized images, and bilevel images such as document and digital halftone images. It is a bitplane coding composed of a new context modeling and adaptive binary arithmetic coding. A target bit to be encoded is conditioned by the estimates of the neighboring pixels including non-causal locations. Several functionalities are also integrated. They are arbitrary shaped ROI transmission, selective tile partitioning, accuracy scalability, and others. The proposed bitplane codec is competitive with JPEG-LS in lossless compression of 8-bit grayscale and 24-bit color images. The proposed codec is close to JBIG2 in bilevel image compression. It outperforms the existing standards in compression of 8-bit color-quantized images.

Recently, a novel approach to color image compression based on colorization has been presented. The conventional method for colorization-based image coding tends to lose the local oscillation of chrominance components that the original images had. A large number of color assignments is required to restore these oscillations. On the other hand, previous studies suggest that an oscillation of a chrominance component correlates with the oscillation of a corresponding luminance component. In this paper, we propose a new colorization-based image coding method that utilizes the local correlation between texture components of luminance and chrominance. These texture components are obtained by a total variation regularized energy minimization method. The local correlation relationships are approximated by linear functions, and their coefficients are extracted by an optimization method. This key idea enables us to represent the oscillations of chrominance components by using only a few pieces of information. Experimental results showed that our method can restore the local oscillation and code images more efficiently than the conventional method, JPEG, or JPEG2000 at a high compression rate.

In this paper, we propose a two dimensional (2D) non-separable adaptive directional lifting (ADL) structure for discrete wavelet transform (DWT) and its image coding application. Although a 2D non-separable lifting structure of 9/7 DWT has been proposed by interchanging some lifting, we generalize a polyphase representation of 2D non-separable lifting structure of DWT. Furthermore, by introducing the adaptive directional filteringingto the generalized structure, the 2D non-separable ADL structure is realized and applied into image coding. Our proposed method is simpler than the 1D ADL, and can select the different transforming direction with 1D ADL. Through the simulations, the proposed method is shown to be efficient for the lossy and lossless image coding performance.

In JPEG2000, the Cohen-Daubechies-Feauveau (CDF) 9/7-tap wavelet filter was implemented by using the conventional lifting scheme. However, the filter coefficients remain complex, and the conventional lifting scheme disregards image edges in the coding process. In order to solve these issues, we propose a lifting scheme in two steps. In the first step, we select the appropriate filter coefficients; in the second step, we employ a median operator to regard image edges. Experimental results show that the peak signal-to-noise ratio (PSNR) value of the proposed lifting scheme is significantly improved, by up to 0.75 dB on average, compared to that of the conventional lifting scheme in the CDF 9/7-tap wavelet filter of JPEG2000.

This paper presents a paraunitary filter bank (PUFB) based on a direct lifting structure of a building block and its inverse transform for lossless-to-lossy image coding. Although the conventional lifting-based filter banks (LBFBs), which are constructed by lifting structures with integer coefficients and rounding operations, suffer from degradation of coding performance due to much rounding error generated by cascading lifting structures, our proposals can be applied to any PUFB without losing many ones because building blocks can be applied to every lifting block as it is. It is constructed with very simple structures and many rounding operations are eliminated. Additionally, the number of rounding operations is reduced more by using two-dimensional block transform (2DBT) of separated transform to each building block. As result, even though the proposed PUFBs require a little side information block (SIB), they show better coding performance in lossless-to-lossy image coding than the conventional ones.

This paper presents an integer discrete cosine transform (IntDCT) with only dyadic values such as k/2n (k, n∈ in N). Although some conventional IntDCTs have been proposed, they are not suitable for lossless-to-lossy image coding in low-bit-word-length (coefficients) due to the degradation of the frequency decomposition performance in the system. First, the proposed M-channel lossless Walsh-Hadamard transform (LWHT) can be constructed by only (log2M)-bit-word-length and has structural regularity. Then, our 8-channel IntDCT via LWHT keeps good coding performance even if low-bit-word-length is used because LWHT, which is main part of IntDCT, can be implemented by only 3-bit-word-length. Finally, the validity of our method is proved by showing the results of lossless-to-lossy image coding in low-bit-word-length.

Set-partitioning in hierarchical trees (SPIHT) is one of the well-known image compression schemes. SPIHT offers an agreeable compression ratio and produces an embedded bit-stream for progressive transmission. However, the major disadvantage of SPIHT is its large memory requirement. In this paper, we propose a memory efficient SPIHT image coder and its parallel implantation. The memory requirement is reduced without sacrificing image quality. All bit-planes are concurrently encoded in order to speed up the entire coding flow. The result shows that the proposed algorithm is roughly 6 times faster than the original SPIHT. For a 512512 image, the memory requirement is reduced from 5.83 Mb to 491 Kb. The proposed algorithm is also realized on FPGA. With pipeline design, the circuit can run at 110 MHz, which can encode a 512512 image in 1.438 ms. Thus, the circuit achieves very high throughput, 182 MPixels/sec, and can be applied to high performance image compression applications.