Systems and methods of denoising images (12) are described. In one aspect, spatially-shifted forward transforms (C1, C2, …, Ck) of an input image (62) are computed. Each forward transform (C1, C2, …, Ck) is computed based on a denoiser transform Z having an associated transpose Z’, wherein a matrix multiplication between Z and Z’produces a diagonal matrix Λ, Z = F(D), F specifies a mapping from coefficients of D to coefficients of Z, and D substantially corresponds to a frequency-domain transform. The forward transforms (C1, C2, …, Ck) are denoised based on nonlinear mappings derived from quantization values linked to the input image (62). Spatially-shifted inverse transforms (C’1, C’2, …, C’k) of the denoised forward transforms are computed. Each inverse transform (C’1, C’2, …, C’k) is computed based on Z and Z’. An output image (40) is computed based on a combination of spatially-shifted inverse transforms (C’1, C’2, …, C’k).

Transformations for denoising images

A data path for a SIMD-based microprocessor is used to perform different simultaneous filter sub-operations in parallel data lanes of the SIMD-based microprocessor. Filter operations for sub-pixel interpolation are performed simultaneously on separate lanes of the SIMD processor's data path. Using a dedicated internal data path, precision higher than the native precision of the SIMD unit may be achieved. Through the data path according to this invention, a single instruction may be used to generate the value of two adjacent sub-pixels located diagonally with respect to integer pixel positions.

Systems and methods for accelerating sub-pixel interpolation in video processing applications

A residual picture is produced and encoded that is a residual picture that is a residual signal between a picture to be coded that is an input moving-picture video signal to be subjected to coding and a predictive picture produced from a reference picture that is a local decoded video signal for each of a plurality of rectangular zones, each composed of a specific number of pixels, into which a video area of the moving-picture video signal is divided. A boundary condition of each of a plurality of borders is obtained between the rectangular zones and another plurality of rectangular zones adjacent to the rectangular zones, and a border, of the reference picture, having a boundary condition that matches the boundary condition, is found by motion-vector search in the reference picture, and border motion-vector data is generated that is data on a motion vector from a border of the rectangular zone in the picture to be coded to the border of the reference picture thus found. A boundary condition of a border that corresponds to the border motion vector data is defined from the reference picture based on the border motion-vector data, and an estimated video signal is generated in each rectangular zone in the picture to be coded, that satisfies Poisson's Equation, thus producing a first predictive picture. The residual picture is then produced with the first predictive picture as the predictive picture and encodes the residual picture.

Moving-picture coding apparatus, method and program, and moving-picture decoding apparatus, method and program

Methods, machines, and computer-readable media for processing an input image with a bi-selective filter in a transform domain are described. In one aspect, a forward transform of an input image is computed. A bi-selective filter smoothes image regions with low magnitude coefficients and sharpens image regions with high magnitude coefficients. An output image is generated from the filtered forward transform.

Bi-selective filtering in transform domain

Disclosed are a method and an apparatus for coding/decoding an image. The method for decoding the image comprises the steps of: deriving a scale factor of a current block, depending on whether the current block is a conversion skip block; and scaling the current block on the basis of the scale factor, wherein the scale factor of the current block is derived on the basis of the position of a conversion coefficient inside the current block, and wherein the conversion skip block is the current block to which conversion is not applied and is specified on the basis of information indicating whether to apply reconversion to the current block.

Method and apparatus for coding/decoding image

We present a new dequantization technique for discrete cosine transform (DCT) based encoding. For the lack of better dequantization methods, the currently existing image compression standards states simply to multiply the received quantized DCT coefficients by the quantization matrix. Therefore, for each DCT coefficient, the quantization noise as large as half quantizer step size may be introduced (in the DCT domain). Our proposed technique, through regularization, dequantizes the quantized DCT coefficients so that the resulting decoded image will be visually more pleasing. Through simulation studies, we show improvements in the visual characteristics and also in terms of the peak-signal-to-noise-ratio (PSNR).

S. Moon-Ho Song

Papers

Regularized dequantization for DCT-based transform coding