Region-Based Segmentation and Auto-Annotation for Color Images

This paper presents an auto-annotation system with simple pre-processed segmentation for digital color image. Recently, annotation techniques become one popular method for image retrieval system in image database management, image recognition system and so on. In the paper, we propose a two-step approach for image annotation. Firstly, the color image is needed to be segmented into two parts: the main object is assumed as the foreground part, and the other will be the background part, and then, only the features of the foreground part (main object) are used for annotation of global images. Here, it is assumed that only a single main object is included in the color image, so annotation problem can be considered as a kind of classification problem, and all the images in database can be automatically categorized by neural network. After image annotating, user can easily retrieve the similar set of images with the same conception that user needs. Finally, we compare the performance of the proposed systems with different processing methods. Experimental results show efficiency of the proposed annotation system.