Assisted Segmentation of High-Resolution Images
There are several techniques of image segmentation in the literature, and the importance of them can be seen in various applications of areas such as computer vision and pattern recognition. Among these areas, interactive image segmentation plays an important role, which uses assisted segmentation algorithms to aplit an image into two regions with the use of markings made by the user. A method which has presented interesting results in this research area is the Superpixel Laplacian Coordinates, which solves this problem with good results based on the quality of the segmentation and with respect to the execution time, even for high resolution images. Although its performance is superior than other algorithms used in assisted segmentation, such a method presents some very costly steps regarding to its execution time, being the most time-consuming step the operation of the SLIC algorithm to generate superpixels. This work aims at finding assisted image segmentation application that can be use in high resolution images. For this purpose, the Superpixel Laplacian Coordinates was used, seeking to optimize its most costly steps and maintain the quality of the final segmentation as close as possible to the result obtained by the original algorithm. A new segmentation technique derived from Laplacian Coordinates was also developed and analyzed in differents sets of images used, called by the present work as Pyramid Laplacian Coordinates. For variantions in the Superpixel Laplacian Coordinates, were considered 4 superpixel algorithms, including SLIC itself, where various input parameters to get a better comparison of their advantages and disadvantages. In addition we implement the parallelization of some parts of the Superpixel Laplacian Coordinates to further reduce its execution time. In addition to finding segmentation methods that can be used in high resolution images, ours work offers a free and open source application that can be used in future works.