用于图像Hash的视觉相似度客观评价测度

由于评价图像Hash性能时,要求对两幅图像是否在感知上相似做出判断,因此针对这一需求,提出了一种衡量感知相似程度的评价测度。该测度的确定是先对图像进行低通滤波,再进行图像重叠分块;然后运用相关系数检测法计算每一对分块的相似程度,并对相似系数归一化,再分别计算若干个最小和最大的归一化相似系数的乘积;最后用最小相似系数乘积与最大相似系数乘积的比值作为感知相似性的测度。实验结果表明,该测度不仅可有效反映图像视觉质量的变化,而且能较好地区分两幅图像是否存在重要的视觉差异,其对感知相似进行评价的性能优于峰值信噪比。

Perceptual Similarity Metric for Application to Robust Image Hashing

To measure perceptual similarity between an original image and its modified version,we propose an objective measure that is relatively stable to normal image processing but quite sensitive to significant changes of the image content in local areas. This is achieved by low-pass filtering the two images,dividing them into overlapping blocks and determining similarity between the corresponding blocks in terms of correlation coefficient that is mapped to the interval \0,1\]. Based on previous calculated correlation coefficients,a ratio is calculated between the smallest and largest correlation coefficients and defined as the perceptual similarity. Experimental results show that the proposed metric is not substantially affected by normal image processing. It provides indication of changes in the image contents when its value becomes less than a given threshold. The proposed metric is useful in applications such as image hashing and CBIR.