Font Size: a A A

Research On Image Superpixel Generation Method And Its Software Implementation

Posted on:2015-07-08Degree:MasterType:Thesis
Country:ChinaCandidate:W YaoFull Text:PDF
GTID:2308330473450834Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
With the development of modern science and technology, the applicable softwares of video image processing have been widely used on mobile terminals, for target tracking, object recognition and so on. Traditional image processing technology mainly focuses on the real-time processing of low resolution images. However, people’s requirements of higher resolution is becoming more and more strict, which makes the traditional image processing technology based upon single pixel unable to meet the real-time requirements of system. Superpixel image generation technique is an image segmentation algorithm by clustering these points with similar properties. It could effectively reduce the amount of data during image processing. As the first step of image and video pre-processing, the quality of superpixel image generation has a great influence on the subsequent image processing. Unfortunately, there is no universal methods to generate satisfactory superpixels for all images. So superpixel generation is still a worthy research topic to further study. This article aims to find a method of generating superpixels.1. Superpixels based on two-step segmentation algorithm is put forward. We first perform preliminary segmentation to get the initial closed area of an image based on the original color image. Then, we do the second division for each large area to get the results.2. In the algorithm of generating closed area, we introduce an active edge growth algorithm to get independent closed area. This algorithm is mainly to find the right point for edge growth through local pixel gradient and color difference to form the closed image edge.3. In the step of further segmentation, we perform dichotomy which divides one large area into two small ones based on the pixel distribution and the color of the area. Then, we merge all the noisy regions based on the similarity between the noise region and its adjacent regions. Finally, we get the results. We carry out a lot of experiments and compare it with other methods. Then, we propose a function to evaluate the superpixels using the statistics of spuerpixel type histogram.4. Finally, we realize the algorithm proposed in this paper through VC++ software platform by designing and cascading some image processing modules.The algorithm of generating superpixels of an image proposed in this paper uses extensive the basic information of color and spatial coordinates. So this method is easy to realize and also can get good results.
Keywords/Search Tags:Superpixels, Edge Detection, Active Edge Growth, Region Dichotomy, Region Merging
PDF Full Text Request
Related items