A data compression device and a compression method are provided. The data compression device includes a quantization table processing unit and a quantization unit. The quantization table processing unit determines a target quantization table in which a quantization coefficient satisfies a data distortion rate and a compression ratio of a preset condition according to a target compression ratio. By constructing different quantization tables for different data, a distortion rate is greatly reduced based on satisfying a compression ratio, and issues that the distortion rate and the compression ratio cannot be simultaneously satisfied in the prior art are alleviated.

A data compression device and a compression method are provided. The data compression device includes a quantization table processing unit and a quantization unit. The quantization table processing unit determines a target quantization table in which a quantization coefficient satisfies a data distortion rate and a compression ratio of a preset condition according to a target compression ratio. By constructing different quantization tables for different data, a distortion rate is greatly reduced based on satisfying a compression ratio, and issues that the distortion rate and the compression ratio cannot be simultaneously satisfied in the prior art are alleviated.

Techniques for coding sets of images with neural networks include transforming a first image of a set of images into coefficients with an encoder neural network, encoding a group of the coefficients as an integer patch index into coding table of table entries each having vectors of coefficients, and storing a collection of patch indices as a first coded image. The encoder neural network may be configured with encoder weights determined by jointly with corresponding decoder weights of a decoder neural network on the set of images.

Techniques for coding sets of images with neural networks include transforming a first image of a set of images into coefficients with an encoder neural network, encoding a group of the coefficients as an integer patch index into coding table of table entries each having vectors of coefficients, and storing a collection of patch indices as a first coded image. The encoder neural network may be configured with encoder weights determined by jointly with corresponding decoder weights of a decoder neural network on the set of images.

Provided is an encoder that achieves further improvement. The encoder includes processing circuitry and memory. Using the memory, the processing circuitry: obtains two prediction images from two reference pictures; derives a luminance gradient value of each pixel position in each of the two prediction images; derives a luminance local motion estimation value of each pixel position in a current block; generates a luminance final prediction image using a luminance value and the luminance gradient value in each of the two prediction images, and the luminance local motion estimation value of the current block; and generates a chrominance final prediction image using at least one of the luminance gradient value of each of the two prediction images or the luminance local motion estimation value of the current block, and chrominance of each of the two prediction images.

Described tools and techniques relate to signaling for DC coefficients at small quantization step sizes. The techniques and tools can be used in combination or independently. For example, a tool such as a video encoder or decoder processes a VLC that indicates a DC differential for a DC coefficient, a FLC that indicates a value refinement for the DC differential, and a third code that indicates the sign for the DC differential. Even with the small quantization step sizes, the tool uses a VLC table with DC differentials for DC coefficients above the small quantization step sizes. The FLCs for DC differentials have lengths that vary depending on quantization step size.

Described tools and techniques relate to signaling for DC coefficients at small quantization step sizes. The techniques and tools can be used in combination or independently. For example, a tool such as a video encoder or decoder processes a VLC that indicates a DC differential for a DC coefficient, a FLC that indicates a value refinement for the DC differential, and a third code that indicates the sign for the DC differential. Even with the small quantization step sizes, the tool uses a VLC table with DC differentials for DC coefficients above the small quantization step sizes. The FLCs for DC differentials have lengths that vary depending on quantization step size.

An image decoding method according to the present document comprises the steps of: receiving a quantized transform coefficient for a target block and a transform index for non-separable quadratic transform; deriving transform coefficients by inversely quantizing the quantized transform coefficient; deriving corrected transform coefficients on the basis of a transform kernel matrix in a predetermined transform set indicated by the transform index; and deriving residual samples for the target block on the basis of inverse linear transform for the corrected transform coefficients, wherein when the target block is divided into a predetermined number of sub-blocks and is coded by intra prediction, the corrected transform coefficients are derived in units of the sub-blocks, and the transform index is received for the target block.

The present disclosure provides a method of reconstructing a video signal based on a reduced secondary transform, which includes: obtaining a secondary transform index from the video signal; deriving a secondary transform corresponding to the secondary transform index, wherein the secondary transform represents a reduced secondary transform, and the reduced secondary transform represents a transform outputting L (L<N) transform coefficient data (L×1 transform coefficient vectors) based on inputted N residual data (N×1 residual vectors); obtaining a transform coefficient block by performing an entropy decoding and a dequantization for a current block (N×N); performing an inverse secondary transform for the transform coefficient block using the reduced secondary transform; performing an inverse primary transform for a block which the inverse secondary transform is applied to; and reconstructing the current block using a block which the inverse primary transform is applied to.

The present disclosure provides a method of reconstructing a video signal based on a reduced secondary transform, which includes: obtaining a secondary transform index from the video signal; deriving a secondary transform corresponding to the secondary transform index, wherein the secondary transform represents a reduced secondary transform, and the reduced secondary transform represents a transform outputting L (L<N) transform coefficient data (L×1 transform coefficient vectors) based on inputted N residual data (N×1 residual vectors); obtaining a transform coefficient block by performing an entropy decoding and a dequantization for a current block (N×N); performing an inverse secondary transform for the transform coefficient block using the reduced secondary transform; performing an inverse primary transform for a block which the inverse secondary transform is applied to; and reconstructing the current block using a block which the inverse primary transform is applied to.