摘要
为了有效地评价图像质量,利用峰值信噪比(PSNR,Pear Signal to Noise Rati-o)和结构相似度(SSIM,Structure Sim ilarity)作为图像质量的描述参数,给出“野点”的定义,提出“野点预测”并基于神经网络(NN,Neural Network)与支持向量机(SVM,Support VectorMa-chines)建立新的质量评价模型:神经网络用来获取质量评价映射函数,支持向量机实现样本分类.采用UTexas图像库数据进行仿真试验,质量评价模型预测图像质量的单调性比PSNR提高7.42%,质量评价模型预测结果的均方误差平方根比PSNR提高36.06%,模型性能测试中“野点”的数目相对减少,模型性能得以提高.试验结果表明该模型的输出能有效地反映图像的主观质量.
Pear signal to noise ratio（PSNR） and structure similarity（SSIM） as two indexes describing image quality were used with neural network（NN） and support vector machine（SVM） to set up new effective image quality assessing model. The definition of isolated points and the prediction of isolated points were illuminated. NN was used to obtain the image quality assessing mapping functions and SVM was used to classify the samples into different types. UTexas image database was used in simulation experiment. With the same level of consistency of quality assessing model, the prediction monotonicity of the model is 7.42% higher than PSNR. The root mean square error （rmse） of the model is 36.06% higher than PSNR. The number of isolated points with the new model was reduced and the performance of the model was enhanced. The results from simulation experiment show the model valid. The output of the new model can effectively reflect the image subjective quality.
出处
《北京航空航天大学学报》
EI
CAS
CSCD
北大核心
2006年第9期1031-1034,共4页
Journal of Beijing University of Aeronautics and Astronautics
关键词
图像质量
支持向量机
神经网络
image quality
support vector machines
neural networking
