基于遗传神经网络的立体图像的客观评价

Objective quality evaluation method of stereo image based on genetic algorithm and neural network

立体图像质量评价已成为立体成像技术领域的关键问题之一,采用基于统计学习的支持向量机模型模拟人类的认知特性对立体图像进行质量评价。但是由于立体图像较单视点平图像数据量成倍增长,为了降低计算复杂度,提取更加符合人类认知特性的图像特征,采用主成分分析提取立体图像样本的特征值和特征向量,利用遗传算法对支持向量机的参数进行最优化选择。实验结果表明:该方法较单纯采用支持向量机方法对立体图像质量进行评价泛化性能更好,其正确分类率达到94%,更符合人眼的主观感受。

The stereo image quality evaluation has become one of the key issues of the three-dimensional imaging technology. This paper is based on statistical learning support vector machine model human cognitive characteristics of the three-dimensional information to evaluate the quality of the stereo image. However, due to the amount of information three-dimensional image is twice the amount of single viewpoint fiat image, in order to reduce the computational complexity, using principal component analysis to extract the eigenvalues and eigenvectors of the three-dimensional image samples, and to use genetic algorithms to select the optimal parameters of support vector machine. The experimental results show that the proposed method has better stereo image quality evaluation generalization performance than by simply using support vector machine method, and has a correct classification rate of 94 percent, more in line with the subjective feelings of the human eye.