ISBN: 978-981-18-1791-5 DOI: 10.18178/wcse.2021.06.010
Integrating Segmentation and Association Relationship for Image Recognition
Abstract— Image segmentation generally refers to partitioning of an image into a set of regions that cover it. It has been believed that these regions may represent meaningful areas of the image, such as buildings, roads, forests, crops, animals and so on. The regions are usually composed of sets of pixels with similar colors. For interesting targets in the foreground, the regions may even take the forms of particular shapes, such as circles, eclipse or rectangles. Inspired from the concept of image segmentation above, it’s interesting to further explore the relationship of segments for the targets of interest according to some association rules, i.e., the spatial association. In this way, we could detect objects from the images satisfying the criterion of both the segment and association relationships. For instance, people may probably prefer to query a scenery image complying with a certain style instead of containing a certain object. In this work, we propose an efficient object detection method integrating information of both segmentation and association relationship. Experimental results indicate that our method has more semantic flexibility for image recognition.
Index Terms— contour discovery, image segmentation, image recognition; association.
Science and Technology on Information System Engineering Laboratory, Nanjing Research Institute of Electronic Engineering, CHINA
Cite: Xin Xu, "Integrating Segmentation and Association Relationship for Image Recognition ," 2021 The 11th International Workshop on Computer Science and Engineering (WCSE 2021), pp. 64-69, Shanghai, China, June 19-21, 2021.