33results about How to "Reduced prediction residuals" patented technology

The invention relates to a vehicle object multi-level knowledge dictionary construction method for monitoring video compression, which is used for expressing and compressing objects in a monitoring video, and comprises the following steps of extracting texture layer knowledge which comprises common textures of different types of vehicles in a multi-source monitoring video, and modeling the texturelayer knowledge by using a texture dictionary; extracting structural layer knowledge which comprises a common structure of the same type of vehicles, and modeling the structural layer knowledge by using a three-dimensional model; and extracting the personalized layer knowledge, wherein the personalized layer knowledge comprises transient knowledge that a certain vehicle individual keeps stable for a period of time, and using a residual dictionary to model the vehicle individual. According to the method, the multi-level knowledge dictionary of the object is established from the three levels ofthe texture layer, the structure layer and the personality layer, the appearance of the moving object in the image in the complex environment can be better expressed, the information such as edge details is richer, the tracking recognition efficiency of the monitored vehicle can be improved, and the method can be widely applied to the aspects of intelligent transportation and the like.

The invention discloses a filtering method for intra-frame prediction. The intra-frame prediction filtering method includes: obtaining filtering parameters; wherein, the filtering parameters include at least one of filtering reference pixels, filtering reference values, filtering ranges, and filtering coefficients; using the filtering parameters to filter the current intra-frame prediction block . Through the above method, the present invention can adjust the filtering method according to the actual situation of the intra-frame prediction block, and select the most effective filtering method for different intra-frame prediction blocks.

The invention discloses a reference frame synthesis method for vehicle object coding in a monitoring video, and the method comprises the steps: generating a global reference frame, generating a localreference frame, and carrying out the fusion of the global reference frame and the local reference frame, and the global reference frame is obtained through the perspective projection transformation of the common information of vehicle objects in a historical monitoring video; wherein the local reference frame is obtained by utilizing local information of a vehicle object in a current to-be-codedvideo, namely the vehicle object in a coded video frame through three-dimensional attitude transformation; and global and local reference frames are fused, a fusion method is designed according to image attributes of different areas of the vehicle, and a final synthetic reference frame is obtained. The local knowledge and the global knowledge are fused to construct the composite reference frame for predictive coding of the vehicle object in the monitoring video, the composite reference frame not only is robust to environmental factors such as illumination, but also contains rich detail information, and the coding efficiency of the vehicle object in the monitoring video is improved.

The embodiment of the invention discloses a video encoding and decoding method, and a video encoding and decoding device. The method in the embodiment of the present invention includes: acquiring the current image, and acquiring the motion vector accuracy of the current image according to the current image, the motion vector accuracy includes a set of search steps, and each search step in the set of search steps is smaller than One pixel unit; obtain the current block and the reference image of the current block; determine the search starting point in the reference image, and search for the target integer pixel point with a pixel unit as the search step from the search starting point; according to the motion Search for the matching block of the current block starting from the target integer pixel with vector accuracy; obtain the motion vector information and the prediction residual signal, and write the information of the reference image, the motion vector information and the prediction residual signal The input stream, the information of the reference image is used to indicate the reference image. The embodiments of the present invention can perform encoding and decoding in combination with video characteristics.

A layered video decoding device (1) includes a parameter set decoder (12) for decoding scale correction information, and a predictive picture generator (1442) for generating a predictive picture based on the scale correction information.

The invention discloses an intraframe prediction performance improvement encoding method. According to the method, prediction sample values are generated by use of mixed four-tap interpolation filtering, including DCT-based interpolation filtering and Gauss interpolation filtering, and for different prediction unit blocks, types of used interpolation filters are determined according to a given threshold; angle prediction mode classification is performed, i.e., intraframe angle prediction modes are classified into three types for carrying out different encoding processing; and boundary prediction sample smooth filtering is performed, i.e., final encoding prediction signals are generated by filtering boundary column or boundary row samples of prediction blocks by use of pixel differences between reference samples. According to the invention, through combination of direction features of intraframe angle prediction and space correlation of the reference samples, the generation process of the prediction signals is optimized, the prediction precision is improved, and better encoding compression performance is obtained.

The invention discloses a prediction-based three-dimensional mesh coding method. In the method, when predicting a normal vector texture image, the relativity between a geometry image and the normal vector texture image and the relativity among three components of the normal vector texture image are fully considered, the quality of the predicted normal vector texture image is high, and thus the predicted residual is small, which facilitates coding and transmission; and the quality of the decoded normal vector texture image is improved greatly, a third-dimension effect of the reconstructed three-dimensional mesh model is upgraded, and users can get excellent visual experience.

The present invention relates to a multi-direction prediction method for block skip scanning in bandwidth compression, comprising: dividing an image into multiple MBs of the same size; and marking the multiple MBs with P marking symbols, wherein P is greater than 1 A natural number; sequentially predict the MBs corresponding to the P marker symbols in a set order to obtain a prediction residual. The present invention determines the predicted pixel value of the current pixel through the method of skipping and scanning. Compared with the existing method, more reference directions can be obtained for MB, and the theoretical limit entropy of prediction can be further reduced.

This application relates to a video encoding method and apparatus, a video decoding method and apparatus, a computer device, and a storage medium. The computer device obtains a to-be-encoded current encoded block in a current video frame and then obtains a first reference block corresponding to the current encoded block in a reference video frame. Next, the computer device obtains, within the reference video frame, one or more second reference blocks matching the first reference block, the one or more second reference blocks and the first reference block being similar reference blocks. Finally, the computer device encodes the current encoded block according to the first reference block and the one or more second reference blocks, to obtain encoded data.

The invention relates to a prediction method of video compression. The method comprises the following steps that (a) a first prediction residual error of an image is calculated via a first predictionmethod; (b) a second prediction residual error of the image is calculated via a second prediction method; (c) a subjective sum of the first residual error of the image is calculated according to the first prediction residual error, and a subjective sum of the second residual error of the image is calculated according to the second prediction residual error; and (d) video compression prediction iscompleted according to the subjective sum of the first residual error and the subjective sum of the second residual error. A more proper prediction method is selected from two methods, and a predictedtheoretical limit entropy can be reduced.

The invention discloses a recursive prediction image compression method based on a gradient direction histogram. The method mainly solves the problems of low grain direction detection accuracy, high time complexity is high, and multiple prediction coefficient numbers. The method comprises the following steps: 1) designing a prototype direction filter; 2) generating a group of direction filter according to the prototype direction filter; 3) reading in original gray level image data and performing grain direction judgement; 4) extending original gray level image data; 5) reading the extended image data by use of a block as a unit; 6) establishing a Markov prediction model according to the read block data; 7) performing prediction compression on the read block data; and 8) judging whether the detection of all block data has finished; ending the compression of the detection of all block data has finished; or jumping a data read address and returning to step 5). By use of the recursive prediction image compression method disclosed by the invention, the grain detection direction can be flexibly and accurately detected, the time complexity is low, the prediction coefficient number is less, the compression performance is improved, and the method can be used for image compression coding of various digital equipment.

The invention provides a video encoding method, decoding method, encoding device and decoding device. The video encoding method comprises the following steps of receiving an original YUV sequence of a video file, conducting a coordinate conversion processing to pixels in the original YUV sequence, acquiring a target YUV sequence and a coordinate conversion parameter, wherein the coordinate conversion parameter comprises a coordinate conversion flag bit for marking a target coordinate conversion method applied to conduct the coordinate conversion processing to the position of the pixels in the original YUV sequence and a target coordinate conversion rule which corresponds to the coordinate conversion method, conducting the encoding to the target YUV sequence and the coordinate conversion parameter, and acquiring encoding bitstream. By the adoption of the video encoding method, coordinate conversion processing is conducted to the pixels in the original YUV sequence of the video file, the correlation of the acquired target YUV in time domain or space domain is stronger, forecasting residual can be reduced, and encoding performance is improved.

The invention provides a nitrogen dioxide concentration prediction method and system, belonging to the technical field of air quality monitoring, and obtains monitoring data such as air pollution monitoring data, meteorological monitoring data, remote sensing reanalysis meteorological field data and geographic covariate data; The forest model, the extreme gradient boosting tree model, and the gated recurrent unit neural network model combined with residual connection process the monitoring data respectively to obtain three predicted values ​​of nitrogen dioxide concentration, and then combine with the weighted average algorithm to calculate the final nitrogen dioxide concentration. Nitrogen concentration value. The invention integrates multi-source spatiotemporal data, and learns the temporal and spatial variation patterns of nitrogen dioxide; through integrated learning combined with the advantages of different algorithms, the stability of the prediction results is improved, the prediction residual error is reduced, and the coverage is wide and the prediction is realized. High-precision, multi-time series nitrogen dioxide concentration prediction; ensures the portability of the machine learning prediction method, and can be directly applied to new monitoring stations with little historical data.