Image retrieval by shape and color

Abstract

The project aims to develop an image retrieval system, which can use digital images as input queries. Relevant feedback approach is also implemented to enhance the retrieval accuracy. This system has numerous potential applications. For example, it can be used as educational software for drawing and recognizing, it also can be used to automatically retrieve trademarks to identify the infringement of similar trademarks. An image retrieval scheme using shape and color features is implemented in this project. Shape and color features are the most important and basic features of an object, so they are used to index and retrieve the object database. For the input, digital images are used. This system has two main functions. First, it can automatically extract the shape and color features from an input digital image and save these features into the database. Secondly, it can retrieve the similar images based on the similarity check between the input query image and the images stored in the database. Only the out most shape is extracted in my project to simplify the implementation of the system. Then the shape is segmented into tokens. Each token is represented by two parameters, local maximum turn angle value and the orientation value. Based on the segmented shape boundary and the centroid of the object, the plane which represents the whole object is divided into token-plats. Then the local color feature of the object can be derived. New similarity measurement algorithm is designed to speed up the image retrieval. Meanwhile, relevance feedback is included in my project in order to find out the most satisfactory retrieval result for individual users. A special version of the system for key retrieval is also implemented. This special version has all the characteristics of the general version. However, it has its particularity designed for retrieving the key.