结合复小波域和空间总变分的无参考图像清晰度评价

No-reference Image Definition Evaluation in Combination with Complex Wavelet Domain and Total Spatial Variation

目的针对频域和空域相结合的清晰度评价方法耗时太长、与主观评价一致性不高等问题,提出结合复小波域和空间总变分的感知清晰度评价方法。方法采用低通滤波的方法将原始图像再模糊,得到其参考图像;在复小波域上采用局部相位一致性计算原始图像和参考图像的清晰度图,在空间域上采用总变分计算原始图像和参考图像的清晰度图;分别将原始图像以及参考图像各自的2幅清晰度图进行几何加权平均,得到2幅整体的感知清晰度图;计算2幅感知清晰度图的相似性,得到原始图像的感知清晰度值。结果在LIVE数据库上的SROCC值达到0.9476,CC值达到0.9483,一幅测试图所耗时间为6.003 s。结论实验证明文中算法的评价性能较优,相较同样结合频域和空域的评价方法计算复杂度降低,耗时较少,更加符合人眼视觉主观感知清晰度。

The work aims to propose the method to evaluate perceptual sharpness combined with complex wavelet domain and total spatial variation with respect to the problem that the definition evaluation method combined with frequency domain and spatial domain is time-consuming and not highly consistent with the subjective evaluation and so on. The resolution charts of original image and reference image were calculated on the complex wavelet domain through local phase consistency. The resolution charts of original image and reference image were calculated on the spatial domain through total spatial variation. The weighted geometric average of the respective two resolution charts of original image and reference image was carried out to obtain two overall perceptual sharpness charts. The similarity of two perceptual sharpness charts was calculated to obtain the perceptual sharpness of the original image. The results showed that the SROCC value on LIVE database reached 0.9476, CC value reached 0.9483 and the test image consumed 6.003 s. Experimental results show that the evaluation performance of the algorithm herein is better than that of the evaluation method similarly combined with frequency domain and spatial domain and it can reduce the computational complexity, consume less time and better conform to the subjective perceptual sharpness of human vision.