Image Information Retrieval

Image Information Retrieval

As the growth of the World Wide Web happen, the popurality in image retrieval has reached to another level. As indicated by a report there are 180 million pictures on the Word Wide Web, an aggregate sum of image information of around 3Tb, and an astonishing one million or more computerized pictures are generating each day. As we try to find the desired image from collection which contains images from various sources such as engineers, teachers, artists and doctors. So image need vary a lot for different groups.

Users search on internet using the different features of images such as shape, texture and color or may use symbolic images. Now more and more sophisticated techniques are evolving which are improving the user involvement with the visual information. So images retrieval system are classified into three broad categories: text based retrieval, content based retrieval and user interactions with image retrieval system.

Text Based Image Retrieval System

Today most of the information Retrieval system are text based. Text based system can give best result in case of general case or in case of specific query at different level of complexity. In the past, images got collected and provided mainly by curator, librarian or by archivist by the use of text descriptors. But as the size of images grows, then textual representation of images become problematic and confusing. For example, images of glass of wine representing Christian mass in reality, may interpreted in different forms.

Image in image retrieval system also associated with its metadata. Metadata consists of attributes like image format, image creator, date of creation or image description taken from titles. But manually doing text attribute assignment are costly and time consuming. But automatic giving the textual attribute using the captions help the blind. Automatic captioning can reduce the cost and time but still many images are still without text.

Content Based Imge Retrieval System

Due to the problem with text based system to access the image, this lead to the path of image based solution. Image based system referred to as content based image system. Content based image retrieval depends on features like shape, color and texture which can easily be gather from the images. Input to the Content based system are of form visual of type image or its attribute. For example, user can input sketch, click on the textual image, or choose the image of required shape. Then system match the given input image with the stored images and output the result. The three methods of content based image retrieval are : Color, Texture and Shape

Texture: The representation of particular texture in image is done by transforming texture into two dimensional gray level variation. The change in brightness of pairs of pixels can help in calculating the degree of contrast, regularity and directionality is calculated. But the problem here is to identify pattern of co-pixel variation.

Color: To identify the image based on color similarity is calculated by making a color histogram for every image. This help in identifying the area in image which are having the specific values. In this area, current effort is to aggregate color proportion by region and by spatial.

Shape: Input for shape is either user sketch a shape or by choosing the image given by the system. The key concept behind shape retrieval system is the identification of features like lines, boundary, circularity and aspect ratio. But the main problem comes when dealing with images with overlapping or touching shapes.

Features like shape, color and texture are crucial visual factor for image representation, but there is still less understanding about the how to combine these attributes in correct ratio. Also understanding the similarity of image retrieval is also important.

User Interaction

User requested images comes from different sources such as education, entertainment, publishing, advertising and art etc. Research in this domain so far, consider only specific collection such as newspaper image archives etc. Experemental result shows that query with visual material shows higher level of specificity than query for textual materials. Certain result also shows that human are highly subjective when it come to aesthetic and emotional needs and this causes problem in case of indexing.

Conclusions

Most of the system is concerned with content based system which mainly focuses on visual attributes such as shape, texture and color. Also there is development of ontologies to further improve the content based system. But there is still some fundamental problems remain such as classification, indexing, user needs, relevance, similarity measures and presentation of retrieval system. But there is clear indication that combining the text and image feature together can improved the system effectiveness. To study image retrieval properly, we need large text collections of images and benchmark queries and to set the evalution measures.