470results about How to "Improve encoding speed" patented technology

The invention provides a method and a device for parallel coding of a panoramic video. The method comprises the following steps of acquiring a spherical panoramic video; mapping the spherical panoramic video into a flat panoramic video according to a preset mapping mode, and segmenting the flat panoramic video into a plurality of plane views; recombining the flat panoramic video into a plurality of video sequences, wherein the plane views in each video sequence are plane views which are mapped to the same position and are formed according to time sequence; performing parallel coding on all the video sequences; restoring decoded videos according to positions of respective decoded videos in the flat panoramic video and splicing the restored videos into a complete flat panoramic video; inversely mapping the complete flat panoramic video into the spherical panoramic video. According to the method, the panoramic video can be coded by fully utilizing multi-core processing capacity of a CPU (Central Processing Unit) / GPU (Graphics Processing Unit), the coding speed of the panoramic video is improved, real-time coding of the panoramic video is realized, and large-scale industrial application and generalization of the panoramic video are favorably promoted.

The invention provides an object and fractal-based binocular three-dimensional video compression and decompression method. In binocular three-dimensional video coding, a left channel is used as a basic layer, a right channel is used as an enhancement layer, and the left channel is encoded by an independent motion compensation prediction (MCP) mode. The object and fractal-based binocular three-dimensional video compression coding method comprises the following steps of: firstly, acquiring a video object partition plane, namely an Alpha plane by a video partition method, encoding the initial frame of a left eye through block discrete cosine transformation (DCT), and performing block motion estimation / compensation coding on a non-I frame of the left eye; secondly, determining the area attribute of an image block by utilizing the Alpha plane, and if the block is not within a video object area of the current code, not processing an external block, and if the block is within the video object area of the current code completely, searching the most similar matching block by a full-searching method in a previous frame of an internal block, namely a reference frame searching window of a left eye video; and finally, compressing coefficients of an iterated function system by a Huffman coding method, and if part of pixels of the block are within the video object area of the current code, and the other part of pixels are not within the video object area of the current code, processing a boundary block independently. The right channel is encoded by a MCP mode and a disparity compensation prediction (DCP) mode, the MCP is similar to the processing of the left eye, and the block with the minimum error is used as a prediction result. When the DCP coding mode is performed, the polarization and directionality in a three-dimensional parallel camera structure are utilized fully.

Through designing encoder for paralleled convolution codes (LDPC), the method carries out encoding, and obtains check matrix H for LDPC code. The encoding procedure is as following: grouped information sequence is stored in memory; through different random interleaving units, the interleaved information sequence is sent to corresponding modulus 2 and processing unit to carry out precoding; then check bit obtained from the precoding is sent to different convolutional coders of recursion to carry out convolutional coding; finally, recombination of check bit outputted from convolutional coders and information bit in memory creates LDPC code words with code rate Rú¢M / (M+N), code length as (M+N)L. Under condition of ensuring performance, the method realizes encoding in high speed. Features are: lower memory space needed, and flexible design and application.

Image pairing information and image classification information are optimized and a Hash code quantization process is used to realize a simple and easy end-to-end deep multi-scale supervision Hash method, and meanwhile design a brand new pyramid connected convolutional neural network structure, and the convolutional neural network structure takes paired images as training input and enables the output of each image to be approximate to a discrete Hash code. In addition, the feature map of each convolution layer is trained, feature fusion is carried out in the training process, and the performance of deep features is effectively improved. A neural network is constrained through a new binary constraint loss function based on end-to-end learning, and a Hash code with high feature representationcapability is obtained. High-quality multi-scale Hash codes are dynamically and directly learned through an end-to-end network, and the representation capability of the Hash codes in large-scale image retrieval is improved. Compared with an existing Hash method, the method has higher retrieval accuracy. Meanwhile, the network model is simple and flexible, can generate characteristics with strongrepresentation ability, and can be widely applied to other computer vision fields.

The invention discloses a high-performance adaptive binary arithmetic encoder which mainly solves the problems of low encoding speed and complicated structure of the existing binary arithmetic encoder. The encoder comprises a probability model storing module, a probability model updating module, an encoding state updating module, a normalization module and register groups, wherein a first register group is inserted between the probability model storing module and the probability model updating module, and a second register group is inserted between the encoding state updating module and the normalization module so as to form a three-level flow structure to improve the encoding speed by sufficiently using the parallelism of hardware; a caching register unit is adopted in the probability model storing module so that the read / write conflict of a probability model memory is effectively avoided; and a prefix zero detection unit is adopted in the normalization module so that the structure of the normalization module is simple and easy to implement with hardware. The high-performance adaptive binary arithmetic encoder disclosed by the invention has the characteristics of high encoding speed and simple structure, and can be applied to the real-time encoding of high-definition videos.

The invention discloses a low-complexity method for selecting HEVC (High Efficiency Video Coding) coding multiple reference frames. The method comprises the following steps of inputting an original coding sequence, taking a code tree cell as the unit, and selecting the optimal reference frame for each coding cell through the operations of a, if the parent coding unit of the current coding unit is in an SKIP mode, taking the optimal reference frame of all forecast cells in the current coding unit as the optimal reference frame of the parent coding unit, and b, if the parent coding unit of the current coding unit is not in the SKIP mode, firstly selecting multiple reference frames for forecast cells in an inter-frame dividing mode of 2N*2N of the current coding unit, and calculating the rate distortion cost of each reference frame in the 2N*2N inter-frame dividing mode, gathering all reference frames of which the rate distortion costs are lower than a threshold in a reference frame set, and selecting the optimal reference frame for the rest inter-frame dividing modes in the reference frame set. The low-complexity method is capable of remarkably speeding up HEVC video coding.

The invention relates to a method for optimizing residual coding by using residual coefficient distribution features and bayes theorem. The method comprises the following steps of (1) reading each frame of images in a video sequence according to a sequence set in a control file; (2) performing intra-frame prediction and inter-frame prediction on the brightness and the chromatic value of each frame of images so as to obtain a residual coefficient; (3) performing texture judgment or bayes model judgment on a residual block TU (the maximum is 32*32) according to the residual coefficient so as to judge whether block encoding is required to be stopped ahead of time or not; (4) performing DCT (discrete cosine transform) and quantification on the whole residual block TU so as to obtain a quantification parameter; and (5) performing entropy encoding on the quantification coefficient, and finally outputting the quantification coefficient in a bit stream mode. By the method, the video encoding speed is increased under the condition that the loss on the encoding quality can be ignored, and videos can be conveniently acquired in real time.

An encoding method by which an encoding speed is improved is disclosed. An encoder (100) comprises an input data storage section (107) for outputting the stored input data (D100) according to an output control signal (108), an input data count section (101) for counting the inputs of input data (D100), an output control section (102) for controlling the output destination of the input data (D100)according to the input counts, one-bit storage sections (103-1 to 103-(N-K)) for holding one-bit data, row-vector storage sections (104-1 to 104-K) for holding row vectors of an LDPC code creation matrix, vector multiplying sections (105-1 to 105-K) for multiplying a row vector and a column vector, a parity storage section (109) for holding a parity created by the multiplication, and an LDPC code-word series creating section (106) for creating an LDPC code word from the input data series and parity series and outputting it.