基于改进图像块分类算法的图像压缩方法  

A Novel Image Compression Method Based on Improved Image Block Classification

作　　者：俞华 韩钰 牛彪[3] 高义斌 赵亚宁 YU Hua;HAN Yu;NIU Biao;GAO Yibin;ZHAO Yaning(State Grid Shanxi Electric Power Research Institute, Taiyuan 030001, China;School of Electric Engineering, Chongqing University, Chongqing 400044, China;State Grid Shanxi Electric Power Company, Taiyuan 030021, China)

机构地区：[1]国网山西省电力公司电力科学研究院,山西太原030001 [2]重庆大学电气工程学院,重庆400044 [3]国网山西省电力公司,山西太原030021

出　　处：《测试技术学报》2021年第4期281-287,共7页Journal of Test and Measurement Technology

基　　金：国家自然科学基金资助项目(51603194);国网山西省电力公司电力科学研究院科学技术资助项目(SGSXDKO0SPJS1900167)。

摘　　要：本文提出了一种基于BP神经网络和改进的图像块分类算法的有效图像压缩方法.首先采用改进的图像块分类算法将图像块划分为互不重叠的3大类图像块,即平滑块、目标块、边缘块;然后基于BP神经网络对平滑块和目标块选用合适的隐含层单元数量进行压缩,对边缘块则采取不压缩而直接保存到压缩数据的方法,最后,得到上述3类图像块压缩数据集的集合.相比于对3类图像块同时进行压缩,该方法相对传统的图像压缩方法节省了0.469 s、峰值信噪比(PSNR)提高了2.11 dB,并使压缩率提高了5.25%,能够更加有效地经过图像压缩后保持细节信息.An effective image compression method is proposed based on BP neural network and improved image block classification algorithm.The improved image block classification algorithm is used to divide the image blocks into three non-overlapping image blocks,namely smooth blocks,target blocks,and edge blocks;based on the BP neural network,the smooth blocks and target blocks are compressed by selecting the appropriate number of hidden layer units,and the edge blocks are saved directly to the compressed data without compression.A collection of three types of image block compression data is obtained.Compared with the traditional method of compressing the three types of image blocks at the same time,this method saves 0.469 s,the peak signal-to-noise ratio(PSNR)increases by 2.11 dB,and the compression rate increases by 5.25%.The ability to maintain image details has been greatly improved.

关 键 词：BP神经网络 图像压缩 图像块分类算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]