Abstract: Aiming at the limitation of correlation method of subjective assessment and its poor effect when facing the problems of local optimum, nonlinearity, higher dimension and small sample, the objective assessment indexes of image quality such as Mean Square Error(MSE), Peak Signal to Noise Ratio(PSNR) and Singular Value Decomposition(SVD), and the scores from LIVE database are used as the learning sample. By using Support Vector Machine(SVM) to train a correlation function between subjective and objective and Genetic Algorithm(GA) for parameter optimization, a perfect model is obtained to achieve the consistency of subjective and objective. Experimental results show that compared with traditional methods, the assessment with the method is more accurate.

Key words: Support Vector Machine(SVM), Genetic Algorithm(GA), image quality assessment, optimal parameter selection, consistency of subjective and objective

摘要： 针对主观评价关联方法易陷入局部最优以及处理非线性、高维、小样本问题时效果不佳等问题，以均方误差、峰值信噪比、奇异值分解这3个图像质量客观评价指标和LIVE数据库评分作为学习样本，通过支持向量机学习得到主客观关联函数，利用遗传算法进行最优参数选取，由此得到具有主客观一致性的评价模型。测试结果表明，相比传统方法，该方法对图像质量的评价更准确。

关键词: 支持向量机, 遗传算法, 图像质量评价, 最优参数选取, 主客观一致性

CLC Number: 

• TP391