Research And Analysis Of Real-time Object-oriented Video Coding Based On CNN

Video information plays a very important position in the multimedia information.Since the video has a very high degree of redundancy, how to provide high compressionratios, while maintaining satisfying picture quality becomes a key to video codingtechniques. Object-oriented approach is a powerful and advanced method for videocoding. The advantage of this approach is that the transmission rate can be greatlyreduced. Since object-oriented coding require massive image analysis, how to solve thereal-time problems caused by processing of large amounts of data operations isparticularly important. Cellular Neural Network(CNN) exhibit outstanding imageprocessing capabilities by virtue of their local couplings, as well as by virtue of theirparallel high-speed processing capability. It is necessary to combine CNN algorithmsand real-time object-oriented video coding, and it is of great academic and practicalvalue to study it.In this paper, the basic concept and development of celluar neural networks isintroduced. Furthermore, the dynamic range and stability of a CNN is analyzed in detail.Next, a CNN-based architecture for real-time object oriented video coding applicationsis analyzed in detail. Then, a new real-time object oriented video segmentationalgorithm based on celluar neural network is presented.The work of this thesis are listed as follows.First, the basic overview of celluar neural networks is introduced, including basicconcept, development and hardware implementation. Furthermore, the dynamic rangeand stability of a CNN is analyzed in detail. All the presentations lay a theoreticalfoundation for the following study.Second, a CNN-based architecture for real-time object oriented video codingapplications analyzed in detail. Simulation results highlight that bit-rate reduction aswell as real-time capabilities are achievable by this approach.Next, a new real-time object oriented video segmentation algorithm based oncelluar neural network algorithms is presented. Through the method of using localadaptive mechanism of celluar neural networks to segmentation and edge detection,process the motion vector field in the video to complete segmention of moving objects.Finally, the results of simulation experiments and performance analyses indicatethat the real-time object oriented algorithm based on CNN can obtain accurate segmentation results and operational efficiency while maintaining real-time system.