Detekce významné oblasti v obraze

Abstract

This work describe implementation of algorithms for searching salient object in image. Program analyzes input image, proceed it through algorithms and save clipping mask, where only salient part of image is highlighted.

We describe step by step creating superpixels, which are atomic part of image. Followed by creating convex hull of point found by Harris corner detector. This give us first coarse imagine of salient object. Mask is finished by counting transmitting energy between superpixels, which distance is counted by Dijkstra algorithm to find shortest path.

Implementation I made in C++ using OpenCV library, which is also free for academic and commercial use. This library include lots of functions, which help us with implementation, and there are optimized for best performance.