Survey Of Image Indexing And Retrieval Computer Science Essay

Published Date: 02 Nov 2017

Professor, Deptt. of Electrical & Electronics Engg., Gautam Buddha University, Greater Noida, INDIA.

[email protected].

Research Scholar, Deptt. of Computer Sc. & Engg., Mewar University, Chittorgarh, Rajasthan, INDIA.

[email protected]

Abstract— Nowadays the world wide web is rapidly growing with the expansion of image database. From the largest amount of image database users have to retrieve relevant images by using an efficient and effective mechanism. Many techniques have come out to fulfill this requirement. One way to provide this requirement is traditional image database indexing and retrieval capabilities in which the image data has to be indexed and to fully convert the image data to an electronic presentation. But there are many factors which prohibit the traditional image indexing including lower quality text and high cost. Problem of traditional image indexing have led to raise the interest in techniques for retrieving images automatically by using intermediate features such as color, shape, texture. This technology referred as Content-Based Image Retrieval (CBIR).

Keywords─ Indexing, Image Retrieval, Image Database.

INTRODUCTION

Images are indexed and retrieved by using textual keywords and visual content. In textual keyword queries, words are used to retrieve relevant from the large amount of image database and in visual queries (CBIR) images are retrieved by using visual pattern characteristics(feature) like color, shape, texture. Image retrieval techniques are based on either text based or content based retrieval (CIBR) have their limitations .Manual indexing(text indexing) of images is highly labor-intensive and may have create problems when having large image database.

The Visual Information Retrieval (VIR) system can use the visual content of the images as indexes example color, shape and texture features [1]. The VIR system is concerned with record retrieval and storage. This system is useful only if it can retrieve the images which have matching capability in real time.

In this system, during input, images are processed to compute the features to represent the image content. The process is known as indexing (or indexation) [2]. In this system we assign a descriptor (keyword) or indices, to each image which will be used by the system in the matching or similarity phase to retrieve relevant or required images and reject the irrelevant one. The indices are stored in database and then the images of the database are ranked according to their matching parameters with the query (fig 1).

indexing of Related images

The main aim of image indexing is to retrieve the similar type of images from an image database using a given query image (i.e. a pattern match).each image has its unique feature. Hence image indexing can be done by comparing their selected feature, which is extracted from the images [3].

An image is indexed by a vector (v) representing the estimate proportion of texture (t).

The procedure for indexing of image is as follows: for each pixel p, compute f (p) that is estimated natural –texture at p .finally v is the proportion of pixel which is classified with texture (t) [1].with the help of this technique we can get the required images. Currently image indexing techniques are of two types.

Textual-indexing (Manual).

Content Based - indexing (automated).

textual -indexing

It is very simple indexing technique. In this technique user acquire keywords which are given for particular image. This technique includes.

Caption indexing.

Classification.

Problem of this technique is that it is:

Labor intensive.

Sometimes inconsistency of text.

content based –indexing

This technique is also known as automated indexing, in this technique images are indexed by using features like color, shape, texture etc. this type of indexing is done by software itself. The image retrieved through this is known as Content Based Image Retrieval (CBIR). There is no problem with this technique like the textual indexing have.

Most of the CBIR indexing techniques based on

Color

Shape

Texture

advantege of cbir

Clear cut analysis.

Extraction process is purely automatic.

Provide compact storage for large image database.

Application area where the CBIR indexing technique is used like

Crime Prevention

The Military

Journalism and Advertising

Medical Diagnosis

Cultural heritage

Education and training

available cbir software

Qbic

It is best known of all image retrieval systems. It offers retrieval by using of features like color, shape and texture and by textual keyword. It is available commercially either in a standalone form or as a part of IBM Product. [4]

Virage

Another commercial system is the VIR image Engine from virage this is available as independent modules. It is easily extend the system by using new types of query interface or customized modules to process collection of images. Application of virage is Alta vista’s AV Photo Finder.

Excalibur

This offers a various image indexing and matching techniques based on company’s own proprietary pattern recognition technology. Its application is Yahoo! Image surfer which allows content based retrieval of images from www [3].

Surfimage

It is an example of European CBIR technology from INRIA, In this system multiple types of images features are combined in different ways and offers relevance feedback facility.

Netra

This system uses features like color, shape, texture and spatial location information which provide region based searching on local image characteristics.

Snapse

This system is an implementation of retrieval by appearances using whole image matching pattern.

Some search engines for search image:-

Picsearch

Ditto

Animation search

Kamat house of picture

Fagan Finder –Image

3 DUP

Alta vista Image search

iconic image indexing

One of the most important problems to be considered in the design of image database system is how to model and access the pictorial data from the image database [7, 8]. The basic aim of Iconic indexing methodologies is the use of pictorial icons as picture indexes.

An image has two kinds of descriptors:-

External information about its content (in textual for).

Internal information related to the shape and the spatial arrangement of its pictorial element.

To make an image database flexible and efficient the spatial knowledge embedded in images should be preserved by the data structures used to store them [6].

image indexing of text

To perform retrieval on text images one way to characterize the document content in a meaningful way. Researchers have addressed a number of problems which is ranging from attempting to identify proper nouns, automatic image abstracting [9]. These types of techniques are appropriate for indexing lower quality text or documents where the recognition results are expected to be poor.

Text characterization

Desilva and Hull [10], have addressed the problem of detecting proper nouns in document images. Proper noun corresponds to the name of people, places and particular object for indexing.

In this approach, segment the document image in to words and filter the proper nouns by examine the words image properties and its relationship to its neighbors. It is very expensive to run contextual post –processing on the recognized image data which may have to identify over half a million names, not to mention the problems associated with word recognition itself [9]. The complexity of Post processing step is reduced by performing pre classification in which possible to additional information from the image that is not available in the recognized text.

The basic idea of providing a characterization and matching /similarity measure is essential for all retrieval problems, including the retrieval of document images.

classical indexing and retrieval

The text which are created electronically have both structured and unstructured components [9]. It is useful to differentiate between document indexes which depend on objective, structured identifiers, such as author names, titles and publishers, and non-objective identifiers which are extracted directly from the text content [9, 12]. If the textual analysis provides structure identifiers, standard database operations can be used to query textual database. And if the un-structured identifiers, method for characterizing the full text content is used in which converted document must be developed.

The basic idea behind many techniques of text indexing and retrieval is to provide the ability to characterize the text corpus in meaningful ways, which allow users to provide a query as a set of keywords, and to provide an effective mechanism to retrieve images in ranked order. One of the most common ways to characterize a document’s image content is to consider the full text document and filter out the words which have negligible amount of effect on the content, and then represent the document by a term vector consisting of the frequencies of meaningful terms. Once the text or documents are indexed, the resulting index vectors can be considered as the signatures and used for retrieval [9]. To query the collection, a simple measure can be used to compute the "distance" between the query vector and the document vector.

document indexing structure

There are two ways to index heterogeneous textual image or document.

Layout Structure (physical).

Semantic Structure (logical).

At the structure level most of the work done is relies on the output of the document analysis process. In a typical document decomposition technique first step will be perform a physical segmentation of document which provides information about physical characteristics. This technique is followed by two type of labeling functional and logical labeling. The functional labeling provides general use of specific physical constructs, and logical labeling provides a document’s semantic components description.

Jaisimha et al. present an overview of text and graphics retrieval system in [9, 13] their system allows keywords searches on raw OCR results but does not provide mechanism to handle degraded documents. After manual segmentation system allows matching of similar images suggest the similar capabilities which are available for signature.

indexing image caption

Indexing image captions is use to establish a relationship between the caption’s content and the image it describes. It has been observed the caption can be a valuable tool, both as a method to retrieve relevant images, and for image interpretation [9].

Srihari make use of association between an image and its caption to attempt to do recognition and Content Based Retrieval [14, 15]. In the PICTION system, she attempt to parse the caption and extract relative location and name information. Candidate’s faces are recognized automatically in the image and for labeling caption information is used. The system is able to differentiate the use of gender in languages as well as spatial relationship.

image retrieval system

In image retrieval we are having a set of reference images and we want to find out the relevant images from this set. The most similar image is the result of the query image which will be retrieved in ranked order (figure 2). We propose the following procedure to achieve this goal [16].

Build the indexing structure over the reference set.

Compute the similarities between a query image and all the reference images.

After this the most similar images are ranked according to computation.

After the third step we can get the relevant images which users want to retrieve.

If we have n number’s of images are indexed effectively by vectors (v1, v2, v3, v4……n) and (v1`, v2`, v3`, v4`………n`) then we define their distance to be [1].

evaluation metrics of image retrieval

In order to evaluate the performance of retrieval system we have to use performance criteria when there was more then one relevant images in the database with respect to a given query,[17] effectiveness of the retrieval in terms of recall and precision method.

The recall R, which means the ability of the retrieval system to retrieve useful images (retrieve all relevant images).

Recall = number of relevant retrieved

Total number relevant.

In contrast, the precision P, this measures the ability to reject irrelevant ones (retrieve only relevant images).

Precision = number of relevant retrieved.

Total number retrieved.

This Recall vs. Precision method is told about efficiency of retrieval (fill ratio).

conclution and futurework

In this paper we have attempted to provide some background past research on both indexing and retrieval. We have summarized the retrieval system ultimately both indexing and retrieval will make use of powerful features offered both in content based and in the underlying content of the text images [9]. These types of systems will need to address the complex tradeoffs between algorithm speed and quality of image.

We are presently addressing the indexing technique. For indexing of image additional efforts will be devoted. To integrate this technique, we have to use Relational Database Management System (RDBMS) to develop new applications. The use of RDBMS allows users a great amount of flexibility of images to be managed concurrently [6]. We are planning to use Oracle for this because it is very powerful and fault tolerant software.