基于视觉特性的多尺度对比度塔图像融合及性能评价

Image Fusion of Multiscale Contrast Pyramid-Based Vision Feature and Its Performance Evaluation

针对同一场景可见光与红外图像的融合问题,提出了一种新的多尺度对比度塔图像融合方法.该方法利用对比度金字塔数据结构得到图像的多分辨序列,采用基于视觉特性的融合算子在图像的相应各级上融合源图像的细节,再通过金字塔逆变换重构出最终融合图像.这种图像处理方法具自适应性,不随各自输入图像的灰度特性而改变,同时增强了融合图像的对比度,产生了较好的视觉处理效果.对熵、交叉熵和互信息3种量化评价标准的进一步分析表明,该方法比传统的Laplacian金字塔、比率低通金字塔和小波变换融合方法的性能更加优化,其中熵提高了0 5%～3%,交叉熵降低了13%～78%,互信息提高了1 8%～8 4%,评价结果与目视效果吻合良好.

Focusing on the fusion problem of the visual and infrared images from same scene, a image fusion method of multiscale contrast pyramid was proposed. By using multiresolution image sequences obtained from the contrast pyramid data structure, the details of source images were fused on each corresponding level with a vision feature fusion operator, then the fused image was achieved through inverse pyramid transform. This processing scheme has adaptability which is unchanged with varying in the global gray level characteristics of each input image, efficiently enhances the contrast of the fused image, and produces better visual effect. According to three quantitative evaluation criteria, it is shown that the proposed method is superior to the traditional Laplacian pyramid, ratio of low-pass pyramid and wavelet transform fusion method, with 0.5% to 3% improvement of entropy, 13% to 78% reduction in cross entropy and 1.8% to 8.4% enhancement in mutual information. The evaluation results coincide with the visual effect well.