A code rate estimation method and a fast mode decision method based on code rate estimation

Abstract

The invention discloses a coding rate estimation method and a fast mode decision-making method based on the coding rate estimation. The code rate calculation formula is as follows: R ori = R hdr +R e where R hdr is the head bit rate, R e is the TU code rate; the calculation of the TU code rate includes: S1) predicting the code rate by the weighted quantization coefficient sum SQC, expressed as R sqc ; S2) Determine the location information Z based on the CG level pos ; S3) Determine the context model inversion parameter M turn ; S4) Determine whether to use the last non-zero coefficient position information Z of the TU level tlp ; S5) Calculate the final code rate of each TU: where R zpos =a 1 Z pos , R turn =a 2 m turn , R tlp =a 3 Z tlp , the R zpos , R turn , R tlp into the formula R e , the calculated code rate is fitted to obtain the coefficient a 1 、a 2 、a 3 . The present invention predicts the code rate quickly and accurately, so as to ensure that the purpose of reducing video encoding burden and saving encoding time is achieved within the range of acceptable performance loss.

Description

technical field [0001] The invention relates to the field of video coding, in particular to a code rate estimation method and a fast mode decision method based on code rate estimation. Background technique [0002] In the video coding process, the mode decision has a huge time consumption. First of all, in the H.265 video codec standard, there is a very flexible block division, such as the quadtree-based coding unit (CU), transformation unit (TU) As well as the prediction unit (PU), all CU / PU / TU need to undergo rate-distortion optimization to obtain the optimal division method. As far as CU is concerned, there are three sizes: 64x64, 32x32, and 16x16. The division is based on the code rate ( The overall performance of R) and distortion (D), which describes the coding quality of the video, the code rate needs to traverse each coefficient to obtain the syntax elements at the coefficient level, in order to complete the entropy coding to obtain the code rate, this process is acc...

AI Technical Summary

Problems solved by technology

[0010] The above code rate estimation technology uses parameters such as the sum of quantized coefficients (SQC), the number of coefficients greater than one (NG1) and the number of non-zero coefficients (NZ) to perform classification fitting after HDQ quantization. The larger the coefficient, the more code rate consumption, so these parameters can intuitively reflect the code rate information, but the disadvantage of this type of method is that the description of the parameters is too rough and does not take into account the correlation between the code rates between the coefficients. Because the code rate of the coefficient is correlated, the code rate of the previous coefficient will have an impact on the code rate of the next coefficient. If it is not considered, it will lead to performance degradation

[0014] The above code rate estimation technology is to obtain the transform coefficients after performing discrete cosine transform (DCT) on the residual coefficients before HDQ, and process and fit the transform coefficients to predict the code rate. This method uses a threshold to imitate HDQ The process, that is, the transformation coefficient judges whether the coefficient is quantized to non-zero through a threshold, and the transformation coefficient is proportional to the quantization coefficient, so the transformation coefficient can also reflect the code rate to a certain extent, but the disadvantage of this technique is that : Because the quantization process is omitted, the use of transform coefficients to describe the bit rate will be more ambiguous, and the context has a tendency to affect the bit rate during the encoding process, while CABAC is encoded based on accurate quantization coefficients, so the use of residual coefficients means Therefore, the correlation between coefficients and coefficient blocks cannot be considered, resulting in serious performance degradation

Images