Image Segmentation Algorithm Based On Otsu And Grab-Cut

Image segmentation,is an important part of image processing and machine vision,and the stand or fall of image segmentation method plays a decisive role in the real-time and reliability of the whole process.In recent decades,people have been looking for different image segmentation methods to solve various engineering problems.By studying the Otsu algorithm based on threshold and Grab-Cut algorithm based on graph theory,aiming at the shortcomings of the two algorithms,this paper puts forward four improved algorithm.Countering the weaknesses of the two-dimensional(2D)Otsu algorithm in the aspects of segmentation quality,anti-noise capacity and running speed,an image segmentation algorithm based on edge detection and Otsu is proposed.By using new area partition method,the proposed algorithm respectively applies three edge detection operators(Sobel/Log/Canny)combined with linear fitting method to limit the object and background between a pair of boundaries which are parallel to the diagonal,and then uses the domain average value of noise pixels instead of the grey value,finally utilizes 2D Otsu oblique segmentation method to separate object from background.Experimental results show that,compared with the current algorithms,it not only has a short operation time but also own a high quality of segmentation,a strong anti-noise capacity and a preferable adaptive ability.To improve the running speed of 2D Otsu algorithm further,a 2D Otsu method based on two boundaries and its fast algorithm are proposed.The proposed algorithm firstly imports Roberts edge detection operator for 2D histogram partitioning,the means of which is the same as the previous algorithm,then constructs the region between the two boundaries into a one-dimensional(lD)histogram and applies 1D Otsu algorithm for segmentation,finally reprocesses noise areas.Experimental results show that,it has not only good segmentation quality and anti-noise capacity but also a faster running speed than the previous algorithm.To improve the segmentation quality of 2D Otsu algorithm segmenting the uneven-illumination image,an Otsu image segmentation algorithm based on double intercept histogram is proposed.This algorithm firstly obtains all the noises and edges on the basis of Sobel edge detection,and replaces the noises and edges into target pixels or background pixels;then according to the image grey distribution divides image into different regions,classifies all regions into two categories,and obtains the line intercept histogram of two categories;finally applies the modified ID Otsu algorithm for segmentation.Experimental results show that,it not only has a good anti-noise capacity and a fast running speed,but also segments conrectly the uneven-illumination image.Aiming at the drawbacks of Grab-Cut algorithm,such as running speed slowly,the bad segmentation quality for the image the targets of which have concaves or thin edges,and Relying on human interaction,An non-interacting Grab-Cut algorithm based on visual saliency is proposed.The proposed algorithm firstly utilizes the SLIC algorithm to segment image into super pixels,then uses a saliency detection algorithm based boundary prior and key regional focus to obtain saliency figure and extract the background seeds,utilizes Min-Cut/Max-Flow algorithm to segment image iteratively by bringing self-adaptive weight coefficient based on saliency figure into the energy function of Grab-Cut algorithm at last.Experimental results show that,it not only has some improvements in running speed and segmentation quality;but also achieves the non-interaction.