Lucent Technologies Inc.

Moushumi Sen

Kyeong Ho Yang

Alireza Farid Faryar

Wenwu Zhu

Methods and apparatus for performing near lossless-concatenated N-level constrained quantization Hadamard transform decoding are provided. Accordingly, the present invention provides a transform-based coding technique, where the coding loss introduced during repeated encoding and decoding is reduced to a substantially insignificant level. This is accomplished by employing a constrained quantization and inverse quantization operation in each level of encoding/decoding in the decoder. In this manner, the concatenation error between each level of a decoder is reduced to a substantially minimum level and the quantization error introduced during quantization operations, i.e., originally at the encoder and in each level of the decoder, is substantially reduced. Thus, the decoded bit stream that is output at the N-th level of the decoder is substantially the same as the original bit stream provided to the input of the encoder.

Methods and apparatus for nearly lossless-concatenated block transform coding

A technique for use in a block transform-based coding system of processing (e.g., encoding and/or decoding) one or more block transform coefficients associated with at least one block of visual data (e.g., an image and/or video sequence) comprises the following steps. First, one or more previously reconstructed block transform coefficients associated with the visual data are identified. Then, a context selection value is computed for use in processing a block transform coefficient associated with the at least one block, the context selection value being based on the one or more previously reconstructed block transform coefficients. The context selection value may be computed as a function of one or more values respectively associated with one or more previously reconstructed block transform coefficients in near proximity, with respect to a scanning order, to the block transform coefficient to be processed. Further, the context selection value may be computed as a function of a spatial frequency associated with the block transform coefficient. Still further, the context selection value may be computed as a function of both the one or more values respectively associated with the one or more previously reconstructed block transform coefficients in near scanning order proximity and the spatial frequency associated with the block transform coefficient.

Methods and apparatus for context selection of block transform coefficients

A technique is provided for use in a video processing system for selecting between an inter-frame coding mode and an intra-frame coding mode when processing a current portion of a video sequence. The technique includes identifying previously reconstructed samples associated with the current portion from a current frame and a reference frame of the video sequence. Then, a selection is made whether to implement the inter-frame coding technique or the intra-frame coding technique on the current portion based on comparisons associated with the previously reconstructed samples. The technique may be implemented in both a video encoder and a video decoder. As a result, the encoder does not need to transmit coding mode-related overhead information to a corresponding decoder. By way of example, the invention may be implemented in a pixel-based coding system or a block transform-based coding system.

Methods and apparatus for context-based inter/intra coding mode selection

A method of perceptually quantizing a block of an image or video sequence includes generating a non-perceptibility of distortion value. The non-perceptibility of distortion value is calculated from one or more masking values, e.g., complexity, brightness, movement, etc., which themselves are respectively calculated from previously reconstructed samples associated with the image or video sequence. The reconstructed samples form one or more causal templates that are used to calculate such masking effects and, thus, the non-perceptibility of distortion value. The perceptual quantization method then includes generating a quantization step size value which is a function of the non-perceptibility of distortion value. Since templates consisting only of previously reconstructed samples are used to derive the quantization step size value, the templates used at the encoder are also available at the decoder. As a result, an encoder does not need to provide quantization-related information to the corresponding decoder since the decoder can get the information using the same causal templates as those used at the encoder.

Methods and apparatus for context-based perceptual quantization

A method of generating a motion vector associated with a current block of a current frame of a video sequence includes searching at least a section of a reference frame to identify previously reconstructed samples from the reference frame that best estimate motion associated with previously reconstructed samples from the current frame associated with the current block. The previously reconstructed samples may form sets of samples. Such sets may be in the form of respective templates. The method then includes computing a motion vector identifying the location of the set of previously reconstructed samples from the reference frame that best estimates the set of previously reconstructed samples from the current frame associated with the current block. It is to be appreciated such a motion estimation technique is performed in both a video encoder and video decoder so that motion vector data does not need to be transmitted from the encoder to the decoder.

Growing community of inventors

West End, NJ, United States of America

Moushumi Sen