Research On Haze Image Saliency Detection Based On Super Pixel

Image saliency detection is one of the hotspots in computer vision and computer image research.It can be applied to content-based image retrieval,image editing,target detection and image segmentation of target objects.Based on the super pixel method and the theory of dark channel dehazing,this paper investigates the saliency detection of haze imagesand proposes a super-pixel based saliency detection method.The proposed approach can be effectively applied to the detection of traffic signs from haze images.The main work is summarized as follows::1)A super-pixel generation method based on regional covariance is proposed.Utilizing the advantages of multidimensional information expression by region covariance,the multidimensional feature information of the image super pixel block is represented as a covariance matrix,and the difference between the feature information of two image blocks is measured by the covariance distance.Firstly,the input image is segmented into several small regions using the K-means algorithm.For each small region,the feature information is described by using the regional covariance matrix.Secondly,the similarity matrix is constructed by using the regional covariance distance between the small area blocks to generate image super-pixels for domain block clustering.Compared with other methods,this method can better maintain the image edge feature information while generating compact super pixels,and improve the image underfill segment phenomenon.2)A dark channel dehazing method based on super-pixel is proposed.Based on the theory of the dark channel,we use the method of region covariance to generate the super pixel block and then replace the dark area of ? region with this super pixel block to estimate the rough transmittance map;Secondly,we use the texture double edge filter and process the transmittance to keep image detail;Finally,we obtain the image after dehazing.The experiment results demonstrate that this method can effectively generates dehazed images with strong reality,while preserving the effective information of the original image.3)The method of saliency detection based on super pixels is proposed.We combine the block of image information and the pixel information to achieve image saliency detection.First,the input image is preprocessed by super-pixel segmentation.Second,the pixel block is calculated based on the regional covariance distance of the pixel block.Finally,the pixel block is sampled to calculate the saliency of the image pixel.The experimental results show that the proposed method avoids the inaccuracy of image saliency detection caused by single noise pixel,and obtains the significant accurate saliency map of the target boundary.Moreover,this paper applies proposed super-pixel based saliency detection to traffic sign detection.First,the input traffic image is dehazed by the the dark channel method based on super pixel;then traffic sign saliency region of the preliminary is determined by saliency detection method based on super pixel in HSV color space;At last,in the saliency region of the image ues Hough algorithm to obtain the region of traffic signs.The experimental results show that the proposed method improves the detection accuracy significantly.