Approaches for Image Database Retrieval Based on Color, Texture, and Shape Features

Approaches for Image Database Retrieval Based on Color, Texture, and Shape Features

Kratika Arora (Sant Longowal Institute of Engineering and Technology, India) and Ashwani Kumar Aggarwal (Sant Longowal Institute of Engineering and Technology, India)
DOI: 10.4018/978-1-5225-2848-7.ch002
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

With an ever-increasing use and demand for digital imagery in the areas of medicine, sciences, and engineering, image retrieval is an active research area in image processing and pattern recognition. Content-based image retrieval (CBIR) is a method of finding images from a huge image database according to persons' interests. Content-based here means that the search involves analysis of the actual content present in the image. As the database of images is growing day by day, researchers/scholars are searching for better techniques for retrieval of images with good efficiency.This chapter first gives an overview of the various image retrieval systems. Then, the applications of CBIR in various fields and existing CBIR systems are described. The various image content descriptors and extraction methods are also explained. The main motive of the chapter is to study and compare the features that are used in Content Based Image Retrieval system and conclude on the system that retrieves images from a huge database with good precision and recall.
Chapter Preview
Top

Introduction

The extensive use of digital images is on rise with each passing day, fuelled by the rapid expansion of digital image content in today’s world of internet. Professionals of different fields are trying to make use of the opportunity to distantly access and manipulate stored images for providing better service to the clients. Professions such as advertising, engineering, law enforcement, fashion designing, graphic designing, medicine, crime prevention, publishing and architectural designing are making extensive use of digital image databases for maintaining record of images and use them when necessary. This leads to demand of a system that can rapidly and effectively retrieve images which are similar and are also relevant. For many years, researchers have been working on the retrieval processes. Hence the image retrieval systems are used for searching, browsing and retrieving images from large image databases. These image retrieval systems can be classified into three categories.

  • Keyword Based Image Retrieval (KBIR)

  • Content Based Image Retrieval (CBIR)

  • Semantic Based Image Retrieval (SBIR)

Keyword Based Image Retrieval uses metadata such as notes, keywords, remarks, annotations, tags, captions, or descriptions linked with the image. These descriptions are man-made and are very subjective. Moreover manual addition of tags in each image is time-consuming, tedious and error prone as image description may vary from person to person. Also, it may not provide the whole information describing the image (Gong et al., 1994). For example, if the image contains an object such as a bus or a car, the user uses his knowledge and adds tags describing the type, the model name, the size and the color. In this case there may be some details which can be missed like object orientation or surrounding objects. Hence the image search initiated using keyword based retrieval system leads to the image retrieval problem wherein the search results may be irrelevant and are not acceptable.

The Content Based Image Retrieval uses an advanced approach compared to Keyword Based Image Retrieval systems (Zhang et al., 2003). In these systems, a query image is given as an input and images similar to the query image are retrieved from the stored database according to persons’ interest. Content based here means that the search involves the actual content present in the image which can be color, texture, shape or any other useful information that can be obtained from the image. In traditional content based image retrieval systems, the visual content of the images in the database are extracted and described by multi-dimensional feature vectors. Here features used can be low level or high level. The low-level features are based on the global image statistics. Since these features are determined for the complete image, they do not separate the items from the background. Hence, they produce good results when queried for the same image or most similar image like pictures having slight variation, modified by adding some text or a frame. When the query demands on finding images containing the same or similar objects, low level features are not sufficient and the high-level feature algorithms are needed. An example of this system is shown in the Figure 1 and 2. Figure 1 is the query image given by the user and Figure 2 is the retrieved images using content based image retrieval system.

Figure 1.

Query image

Figure 2.

Retrieved images of content based image retrieval system

Complete Chapter List

Search this Book:
Reset