Research On Bottom-up Visual Saliency Detection Algorithms

Scientific research shows that visual saliency detection can quickly detect the interested information from the amounts of image data and improve the efficiency by simulating the human visual system.In recent years,the visual saliency detection technology has been widely used in image retrieval and classification,object segmentation and recognition and other fields.It is still a hot topic to design the visual saliency detection method in the rapidly development of digital image processing world.Human visual system scans the image in both bottom-up,rapid and task-independent manner and top-down,slow and task-dependent manner.So far,most visual saliency detection method is bottom-up.This paper aims to study the bottom-up visual saliency detection and is organized as follows:A new visual saliency detection algorithm based on wavelet transform and simple priors is proposed.Firstly,the color space is converted by wavelet transform for each color channel.Then,the wavelet coefficients are used to generate the reconstruction feature maps and the local contrast method is used to calculate the local saliency.In order to highlight the rarity in the whole image,a probability density function with normal distribution in multi-dimensional space is utilized to calculate the global saliency.Finally,the algorithm also introduces two simple priors(color prior and center prior)to reduce the saliency of background region and enhance the robustness.The experimental results show that the proposed algorithm can effectively detect saliency area and be superior to the existing transform domain algorithm.Visual saliency detection based on multi-scale and Markov chain is presented.Firstly,the input image is segmented by superpixels and construct sparse graph.Then,the superpixels nodes of the boundary on the graph are chosen as the absorbing nodes and the other superpixels nodes are treated as transient nodes in a Markov absorbed chain.Then,the transition probability matrix of the random walk in Markov absorbing chain and the absorbed time from each transient node to absorbing nodes are calculated.Furthermore,the absorbed time is normalized to get the saliency value and the saliency map.In order to suppress the background and highlight the saliency regional,the absorbed time is weighted and constrained and is extended to multi-scale.In order to demonstrate the effectiveness of the proposed algorithm,it is compared with the existing outstanding bottom-up saliency detection model.The experimental results show that the proposed saliency detection algorithm can effectively obtain saliency regional of the image and perform better in suppressing background than other models.