Research And Implementation Of Image Segmentation Algorithm For Video Encoding And Decoding Scenarios

At present,video encoding and decoding are widely used in the field of video images,and at the same time,enormous computing power is required for the encoding and decoding of video images.Therefore,computing power has become an important factor affecting video encoding and decoding efficiency.In the process of video encoding and decoding,due to the disturbance of the external environment,non-target objects,such as the background,are often subjected to useless encoding,which wastes a lot of computing power.Domestic researches on video encoding and decoding efficiency mainly focus on the optimization of the encoding block during intra-frame or inter-frame prediction after the encoding block is generated,but they do not involve the optimization of the video image before the generation of the encoding block.In view of this situation,an image semantic segmentation algorithm for video encoding and decoding scenarios is proposed here.Here,from a new optimization perspective,a new image segmentation algorithm is used to generate encoding blocks,and the non-target parts such as the background are segmented in advance,so as to avoid useless recoding of the background and other non-target objects during the encoding process.In this way,the computing power consumed in the encoding and decoding process is reduced and the encoding efficiency is improved.The main work of the thesis includes:(1)Propose a new image segmentation algorithm.This algorithm divides the image into n×n encoding blocks to realize the segmentation of the image in units of encoding blocks.The segmentation of each pixel in the image is transformed into the segmentation of each encoding block in the image.By segmenting the background and other non-target objects in the image in advance,annotated encoding blocks are generated,and the optimization of video encoding and decoding is realized by optimizing the generation of encoding blocks.The problem of encoding block generation is transformed into the problem of image segmentation processing.(2)Realize the image segmentation algorithm based on convolutional neural network.Convolutional neural network has a powerful feature extraction function.The use of convolutional neural network can automatically make up for the defects of traditional algorithms due to the limitation of prior knowledge,extract the features of the video image,and generate encoding blocks based on the image features.(3)Optimize the neural network according to the hardware characteristics.According to the characteristics of hardware channel alignment,the neural network structure is optimized to accelerate the training process of the neural network under the GPU and the inference process under the hardware MLU.Through the above three optimizations,the non-target parts of the video image are segmented based on the image segmentation algorithm,and the encoded blocks required by the video encoding and decoding are generated.The tagged encoded blocks obtained by the segmentation can reduce the useless encoding in the video encoding and decoding.Finally,through the convolutional neural network and its optimization,the training and reasoning process of the neural network is accelerated,and the video encoding and decoding technology is optimized.