Indian Journal of Science and Technology

Abstract

Objective: To develop a system that will optimize storage, automatically classify and tag images using image content and image context by using classification techniques and construct image archive to enable enhanced reference and inference with a simple query. Methods: The approach uses both image processing and data mining techniques. Multi modal processing of the Human brain has the ability to distinguish the components in an image and understand the content instantly. To automate this processing ability by the Computer, Computer Vision attempts various techniques. Features that describe the objects of interest are extracted by converting the image into Hue, Saturation, Value (HSV) space. Further segmentation, profiling and thresholding techniques are applied. Finally K Nearest Neighbours (KNN) classifier has been used for classifying the images. Hierarchical decision tree has been built by uniquely identifying classes that could be used for tagging. Findings: Two levels of classification have been addressed in this paper. Statistical features are extracted from the H-plane of the HSV color space and with an appropriate threshold obtained from the histogram, images have been classified into indoor and outdoor images at the first level. Further the indoor images are classified using segmentation and profiling, as classes of images with presentation and without presentation. Applications/Improvements: Images can be segmented based on more image content that signify the event by watermarking technique and by extracting more features. These segments could further be used to build the decision trees that will identify distinct classes. The depth of the decision tree will enhance indexing, tagging and improve the efficiency of query based image retrieval.
Keywords: Content Based Image Retrieval (CBIR), HSV, KNN, Region Based Image Retrieval (RBIR), Region of Interest (ROI), Threshold