基于二进神经网络的稀疏编码架构的图像无损压缩新方法

视频图像的高效无损压缩在海量的航空和遥感图像传输、珍贵的文物信息的保存等方面具有重要的应用价值,而目前的研究热点主要针对有损压缩,为此通过对现有的无损压缩方法的分析和研究,提出一种在稀疏编码与二进神经网络相结合的框架下建立新的图像无损压缩方法.首先借助二进神经网络中的线性可分结构系建立冗余字典,获得有效的稀疏分解基;再借助二进神经网络学习算法将图像映射为以线性可分结构系为神经元的二进制神经网络,在此基础上建立相应的模式匹配算法将每个神经元与冗余字典简历映射关系,通过稀疏系数建立原始图像的编码形式,进而实现了图像的无损压缩,并从理论上分析了该方法可以有效地提高压缩比,最后通过实验验证了该算法的有效性和通用性.

A novel lossless image compression scheme based on binary neural networks with sparse coding

The high efficient lossless compression of videos and images has important application value in the transmission of massive aerial and remote sensing image, the storage of precious cultural relics, and so on. However, the current researches mainly focus on lossy compression. So, according to the analysis and research of the existing lossless compression methods, a new method for lossless image compression based on the combination of sparse coding and binary neural networks is proposed. Firstly, a redundant dictionary is constructed by using the linear separable structure of the binary neural networks to obtain efficient sparse decomposition. Then the binary neural network learning algorithm is used to map the image into a binary neural network with linear separable structure. Based on this, the corresponding pattern matching algorithm is constructed to map each neuron to the redundant dictionary resume. Also the coding of the original image is established by the sparse coefficients. Finally, the lossless compression of the image is realized. The theoretical analysis shows that this method can effectively improve the compression ratio. Experiments verify the validity and generality of the algorithm.