Design And Hardware Implementation Of Lossy Reference Frame Compression Algorithm For Low Power H.264 Coding

In recent years,with the development of multimedia and Internet of things,H.264 encoders are widely used in video surveillance systems,wearable devices and so on.While the performance of H.264 encoder is increasing and the power consumption continues to decrease,the access power of the encoder is increasing during inter-frame prediction,it occupies about 60%or more of the total system power consumption.Compressing the reference frame can reduce the amount of data stored,thereby reducing the access power.In different compression techniques,lossy compression can reduce large amount of stored data and becomes a hot topic of research.Firstly,a variety of lossy compression algorithms were introduced,some standard test pictures were used as benchmark to verify these lossy compression algorithm,it is concluded that the I-ADPCM algorithm has the best performance under the same compression ratio.In order to solve the problem that predictive model for I-ADPCM algorithm can not predict complex image,for the Y component,seven kinds of 4 x 4 small block directional prediction mode and one kind of non-directional prediction mode are added so that the prediction error is reduced.At the same time,a new model selection method is proposed to improve the hit rate of the best model with the feature that combing the principle of minimum quantization and minimum residual.In hardware design,the pipeline design and parallel design were carried out for both encoder and decoder,so that the encoding system can complete the 480p @ 30fps encoding without increasing the clock frequency.In order to minimize the power consumption of the encoder and decoder,a fine-grained clock gating strategy was presented both at the system level and SRAM level,the power consumption is reduced by 15.4%for the encoder and 9.6%for the decoder.At the end of this paper,the function and power consumption are analyzed.In the case of reference frame compression ratio of 50%,the average signal to noise ratio of Y component is reduced by 0.204dB.In the case of coding rate 480p/30fps,the amount of data accessed by the reference frame is reduced by half,H.264 coding system's SDRAM power consumption down 44.2%,and the whole coding system energy consumption is 3.9nJ/pixel.