基于VGG19卷积神经网络的图像拼接质量评价算法

Image Mosaic Quality Evaluation Algorithm Based on VGG19 Convolutional Neural Network

针对图像拼接质量评价算法多数存在没有分析拼接前后图像的综合图像像素信息和结构信息的问题,本文提出了一种基于VGG19深度卷积神经网络的图像拼接质量评价方法。采用VGG19网络,提取拼接图像和原图像的卷积特征,分别计算2个图像特征图的Gram矩阵,并以2个Gram矩阵的差异作为评价图像拼接质量的指标。同时,为了验证图像拼接质量评价算法的可行性,选取同一场景下的5幅图像进行实验测试。测试结果表明,L s函数值越小,图像的拼接效果越好;而随着L s函数值的增大,图像的拼接效果逐渐变差。说明该算法的评价结果符合人眼的主观评价,能够有效评价图像的拼接质量,可以作为评价同一场景下图像拼接效果的有效指标。该研究有效解决了双目视觉下自动导向车(automated guided vehicle,AGV)在2个摄像头图像拼接处的质量评价问题,具有一定的创新性。

In view of the problem that most of the current image stitching quality evaluation algorithms do not analyze the comprehensive image pixel information and structure information of the image before and after stitching,this paper proposes an image stitching quality evaluation method based on VGG19 deep convolution neural network.VGG19 network is used to extract the convolution features of the stitched image and the original image,and then calculate the Gram matrix of the two image feature images respectively.The difference between the two gram matrices is used as an index to evaluate the quality of image mosaic.Experiments show that the evaluation results of this algorithm accord with the subjective evaluation of human eyes and can effectively evaluate the quality of image mosaic.