CBIR System Using Scaled Conjugate Gradient Feed-Forward Neural Network

Abstract

Images produced by copious of application has now become the centre of attraction for researchers. The task of retrieval of desired image from the ocean of images is quite tedious task. The paper focuses on mining of database from the picture descriptions that are recovered from the same picture known as Content Based Image Retrieval (CBIR). In this paper, we have used colour, texture and shape picture description and Scaled Conjugate Gradient feed-forward neural network (SCG-FFNN). Initially we have prepared the model using colour histogram for colour feature, wavelet moments for texture feature and edge-histogram and edgedirection for shape features; then trains the dataset using FFNN. The trained database is tested with a picture, and generates the like pictures from it. The execution of the system is calculated with the help of precision and recall parameter and the outcomes are found to be high as compare to other research effort. The practical work has been conducted with Corel and Caltech dataset.