基于小波变换和Retinex算法的显著性检测

Saliency detection based on wavelet transform and Retinex

针对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,提出了一种基于小波变换和Retinex算法的显著性检测算法。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,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。

Based on the saliency maps gener adted by many previously proposed algorithms had many issues such as background information,more noise,losing a significant number of characteristics. To put forward a new solutions for the described issues,this paper introduced a wavelet transform and a Retinex algorithm. It applied the Retinex algorithm for the image preprocessing and then processed the image again using SLIC. It generated a feature map of the low-frequency part and the highfrequency part respectively by wavelet transform and obtained the saliency maps by using the bilateral filtering. Finally it generated the final saliency map by weighted combination of the two saliency maps. Experimental data confirms that the proposed algorithm had a small influence in the background,less noise and more characteristics in the saliency map.