Introduction to Seminar Topic on Image Retrieval Using Image Segmentation :
The proposed algorithm for image retrieval has been designed to retrieve the image using the constituent segments within a defined discrete channel. The feature of a particular segment can be used to calculate its distance from another segment while the matching process is done.
The design basis has two approaches namely global feature based image retrieval which is more objective and local feature which overcomes the shortcomings of the former.
Content based image retrieval system
After homogeneous segregation of the image and automated segmentation the features of the image are stored in database so that it can be utilized in CBIR.
Image segmentation and feature extraction
The algorithm used here is unique and denoted by K-means. Through mathematical equations we can draw a relation between K and number of pixels, intensity and number of iterations. Such an algorithm broadly differentiates the bright spots, noise and speckles.
The three parameters considered here are color, texture and shape. Texture features can found by Euclidean distance and other coefficients. The shape is determined by Fourier shape descriptors.
Segment based image retrieval
Here the image goes through a process of segmentation and then the dominant component is highlighted. A relation is established between the weight of the image and the shape features. An experimental setup infers that color and shape are given equal weight and image retrieval is done.
Retrieval algorithm can be improved using the process of segmentation. The scope of algorithm appears to be good. Examining the k-means step it is expected the system of producing classes can be automated. The enhancement of the algorithm would be to add texture while classifying the segments. The experiments also conclude that a variety of images with peculiar features will enable us to setup an assessable image retrieval system.
Download CSE Latest Seminar Topic on Image Retrieval Using Image Segmentation .