证据理论和改进神经网络相融合的图像识别算法  

Image Recognition Algorithm Based on Evidence Theory and Improved Neural Network

作　　者：王丽艳[1] 

机构地区：[1]重庆邮电大学移通学院,重庆401520

出　　处：《微电子学与计算机》2013年第2期148-152,共5页Microelectronics & Computer

摘　　要：针对单一特征图像自动识别算法存在识别结果不稳定和识别正确率低等缺陷,提出一种基于证据理论和改进神经网络相融合的图像自动识别算法.首先提取能反映图像类别信息的颜色和纹理特征,然后采用RBF神经网络对单一特征进行初步识别,识别结果作作为证据,最后采用证据理论对初步识别结果进行决策融合处理,得到图像最终识别结果.仿真测试结果表明,该算法的平均识别正确率达到92.29%,相对于单一特征识别算法,图像识别结果的可靠性和正确率得到了大幅提高,具有较好的应用前景.Because single feature automatic image recognition algorithm＇s identification result is not stable and the correct recognition rate is low blemish, this paper put forward a image automatic recognition algorithm based on based on evidence theory and improved neural network. Firstly, color and texture features of image are extracted which can reflect the image category information, and then RBF neural network is used for single feature identification and recognition results are taken as evidences, the evidence theory is used to fuse the identification results and gets the final recognition result o~ image. The result of the simulation shows that the proposed algorithm＇s the average recognition rate is up to 92. 29%, and compared with the single feature recognition algorithms, image recognition results of reliability and accuracy is increased greatly, and it has better application prospect in image recognition.

关 键 词：颜色特征 纹理特征 神经网络 证据理论 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]