基于小波变换和Retinex算法的显著性检测

近些年来，关于图像显著性检测的研究越来越热门。基于之前提出的很多算法产生的显著图都存在背景信息杂乱、干扰噪声多、细节丢失等问题，本文提出了一种基于小波变换和Retinex算法的显著性检测算法来解决以上问题。首先，利用Retinex算法对图像进行前期处理；然后，对前期处理过的图像进行SLIC超像素分割，对超像素进行小波变换，分别生成原始图像低频部分和高频部分的特征图，并进行适当的双边滤波降噪，生成对应的显著图；最后，通过加权组合这两种显著图，得到最终的显著性图。实验结果表明，本文提出的算法生成的显著图具有受背景影响小、噪声少以及细节突出等优势。

Saliency detection based on wavelet transform and RetinexSun Qingwen,Qiu Weigen

Saliency detection have become a highly active research field in the recent years. Based on many previously proposed algorithms, saliency map produces background information clutter, more noise, a significant number of characteristics are lost and many other issues. This paper proposes new solutions for the described issues, it introduces a wavelet transform and a Retinex algorithm. It applied the Retinex algorithm for the image pre-processing then processed the image again using SLIC. It computed the output to a wavelet transform which generated a feature map of the low-frequency part and the high-frequency part respectively, it obtained the saliency map by using the bilateral filtering then it generated the final saliency by a weighted combination of the two saliency map. Experimental data confirms that the proposed algorithm have a small influence in the background, less noise and more characteristics in the saliency map.