Object-Based Image Content Characterisation for Semantic-Level Image Similarity Calculation

Abstract

This paper proposes an approach to semantic-level image similarity calculation for object-based image retrieval. It is not only suitable for images with single objects, but also for images containing multiple and partially occluded objects. In this approach, contours of objects are used to distinguish different classes of objects in images. We decompose all the contours in an image into segments, and compute features from the segments. The C4.5 decision-tree learning algorithm is used to classify each segment in the images. Each image is represented in a k-dimensional space, where k is the number of classes of objects in all the images. Each dimension represents information about one of the classes. The class information about each class can be obtained through either summing up the number of segments that belong to the class, or summing up the probabilities that each segment belongs to the class. Euclidean distance between images in the k-dimensional space is adopted to compute similarities between images based on either class predictions of segments or probabilities of segment classes. Experimental results show that this approach is effective.