摘要:
A device may be configured to compress feature data according to one or more of the techniques described herein. In one example, feature data may be compressed by using residual encoding to enhance the feature data by removing redundancies. Enhanced feature data may be spatially down sampled and the number of channels of the enhanced feature data may be reduced by applying a 2D convolution operation. A heatmap based on the reduced enhanced feature data may be generated. The reduced enhanced feature data may be scaled using the generated heatmap. The scaled reduced enhanced feature data may be entropy encoded to generate a bitstream.
摘要:
A method of decoding a video through symbol decoding, the method including parsing symbols of image blocks from a received bitstream; classifying a current symbol into a prefix bit string and a suffix bit string based on a threshold value determined according to a size of a current block; performing arithmetic decoding by using an arithmetic decoding method determined for each of the prefix bit string and the suffix bit string; and performing inverse binarization by using a binarization method determined for each of the prefix bit string and the suffix bit string.
摘要:
A method and device for processing LDR images of a video sequence to improve image quality. The method comprises temporally decomposing successive HDR frames of a video sequence and the corresponding LDR frames and performing a comparison between the HDR and LDR frequency sub-bands. A current LDR image can then be modified on the basis of a comparison between the frequency sub-bands.
摘要:
Provided is a video encoding method, in order to encode a current region of a video, performing transformation on the current region by using transformation units in a variable tree-structure which are determined from among transformation units that are hierarchically split from a base transformation unit with respect to the current region and which are generated based on a maximum split level of a transformation unit; and outputting encoded data of the current region, information about an encoding mode, and transformation-unit hierarchical-structure information comprising maximum size information and minimum size information of the transformation unit with respect to the video.
摘要:
Methods of data encoding using trees formed with logic gates are described which lead to spatial compression of image data. Data encoding is achieved using a five-level wavelet transform, such as the Haar or the 2/10 transform. A dual transform engine is used, the first and engine being used for the first part of the first-level transform, the second part of the first-level transform and the subsequent-level transforms being performed by the second transform engine within a time interval which is less than or equal to the time taken by the first transform engine to effect the part-transform. Each bit plane of the resulting coefficients is then encoded by forming a tree structure from the bits and OR logical combinations thereof. Redundant data are removed from the resulting tree structure, and further data can be removed by using a predetermined compression profile. The resulting blocks of compressed data are of variable length and are packaged with sync words and index words for transmission so that the location and identity of the transformed data blocks can be determined from the received signal.
摘要:
The invention relates to a method and device for mixing video streams in a video mixer device, by means of which a plurality of input video streams from different subscribers which are encoded with code words for macroblocks and in which the code words have interdependencies are combined into an output video stream. The input video streams are at least entropy-decoded to such a degree that the dependencies among the code words are dissolved, wherein the macroblocks are re-organized and mixed with each other, and the mixed macroblocks are entropy-encoded to obtain a new dedicated video stream.