Design and Development of Content Based Image Retrieval System using Machine Learning

Abstract

In the current era the storage, retrieval and transmission of digital images in multimedia applications is used by various human activities. The major challenges of digital imaging research are to obtain target image for user queries. This can be solved by either using text annotation for the image or to formulate a method for rendering an image based on the content of the visual image, which is called Content Based Image Retrieval (CBIR). In the former approach categorization of images are done manually, which is inappropriate if the database is dynamic and huge. Also text annotation done by human can be ambiguous and subjective that causes complexity of queries. In the first part of our study, we discuss the problem of database access based on images in an index considered from the contents of the image itself. Various sets of models are stored in the database using low-level features i.e. shape, texture, and color, and similar categories of images are retrieved based on the query images. Color is a relatively robust visual feature, independent of the size and resolution of the image, or its orientation. The similarity between two images given will be calculated based on the comparison made between the color histograms of the images in the database and that of the query image. Texture of an image corresponds to a perceptual phenomenon, easily detectable by humans, but difficult to describe mathematically. Texture is a measure of the repetitive elements in the image. Shape is a key feature of image region separation, and its efficiency and flexibility play an important role in retrieval. The shape of an image can be divided into two groups: trans-boundary and regional. The former uses the outer bounds of the shape, while the later is based on the entire area of the shape. The two most popular representations for these two groups are Fourier descriptor and invariant moments. newline