International Business Machines Corporation

Nec Electronics America, Inc.

Nader Mohsenian

Charles E Boice

Agnes Yee Ngai

Adrian Stephen Butter

Robert L Woodard

Brian J Cascarino

A de-interlacing architecture is taught. The de-interlacing architecture adopts a perceptual model to measure membership probabilities for a collection of image samples of an interlaced video source with respect to extracted static, motion, and texture image components of the same collection. The probabilities are used to prioritize contributions from the three image components and produce a progressive video sequence which is a summation of the portions of the aforementioned components. The perceptual model uses a dual-stage motion-based image difficulty measuring scheme to equalize contributions from the three image components in a manner that video artifacts in the output signal are least perceptive. A parameter mapping technique composed of several logic units, a decision function, a weight assignment block, and a look-up table, will be presented to derive the final component weights. The mapping technique contains a multitude of thresholds and decisions which aid in interpolating the missing lines of the progressive frame.

Video data de-interlacing using perceptually-tuned interpolation scheme

A frequency domain data management technique for producing pseudo-constant bits per picture compressed video bit-streams in a low delay digital encoding environment is presented. This technique forms a hierarchy among the localized samples of the picture in terms of frequency importance and the picture difficulty after a shot-change is detected. After a shot change, the data management technique implements a series of tasks composed of picture difficulty evaluation, frequency classification, frequency constraining, and zero bytes generation to achieve a pre-determined average picture bits. Further, the low delay encoder uses a unique updating mechanism to encode certain regions of the pictures in intra mode and ensures that the whole picture is updated after a pre-selected number of pictures. The updating method disseminates compression artifacts throughout the video stream by changing the orientation of the intra-coded regions for every picture and scatters intra-picture compression artifacts by spatially decimating the aforementioned regions at different rates.

Method and apparatus for producing pseudo-constant bits per picture video bit-streams for low-delay compression system

A real-time single pass variable bit rate control strategy is provided to achieve variable bit rate (VBR) MPEG-2 encoding in a video compression system. For a sequence of frames, the level of encoding difficulty of a video interval (i.e., a group of pictures (GOP)) is determined by a perceptual rate-quantization ({overscore (C)}-{overscore (Q)}) model. This model assigns a composite ({overscore (C)}-{overscore (Q)}) curve to each video interval from which the number of bits for the video interval is estimated. The estimation relies on a causal predictive model using the parameters obtained from previously encoded video intervals. The R-Q relationship of each picture type is updated and picture bits are assigned, based on the total rate of the video interval under analysis. Robustness of the variable bit rate control strategy is ensured throughout scene transitions and instabilities by applying, for example, a non-linear median filter, and a low pass filter, respectively.

Real-time single pass variable bit rate control strategy and encoder

Method, system and computer program product are provided for adaptively encoding in hardware, software or a combination thereof a sequence of video frames in real-time. Pre-encode perceptual activity measurement processing is employed to derive statistics on each frame of the sequence of video frames to be encoded. The statistics are used by variable bit rate logic to obtain a number of bits to be used in encoding each frame. The number of bits to be used is provided to a single encoding engine, which encodes the sequence of video frames and produces a constant quality, variable bit rate bitstream output. The pre-encode processing employs a regulator as the global data flow control and synchronization for the encoder. Perceptual activity analysis on each frame of the sequence of video frames can derive information on, for example, shading, scene change, fade, color, motion and/or edge presence within the frame. Voting gives greater weight to the presence of certain characteristics within the frame.

Growing community of inventors

Plainsboro, NJ, United States of America

Nader Mohsenian