Description

Content Based Image Retrieval is the process of identification of similar images present in the database. A Feature based approach is employed for the identification of the similar images from the dataset. The features were extracted from the images based on three different feature extraction techniques like color, texture and shape based features. Inorder to extract color features Color Histogram features, color moments features and color correlogram features were extracted. Shape features were extracted based on edge detection in the input image. Inorder to extract shape features canny, sobel and prewitt operators were employed. The combination of features will improve the performance of the process. The performance of the process is measured in-terms of Precision and Recall. The input color images were collected from the image retrieval database. The three type of features were extracted from the images which is very large in dimensions. Best features were selected from the test features based on Neighborhood Discriminant Hashing (NDH) process. NDH estimates the probability of selection of each features in the train features. The features that were having greater counting of values greater than half of the estimated probability were identified as the selected features. The selected test feature and train feature were used for image retrieval process. Image retrieval process is employed based on classification approach. The classification of the data is done based on Probabilistic Neural Network (PNN). PNN is a feed forward neural network that recognizes patterns in the input images. The category of the test image is identified similarly the selected train features were classified and the category of each image in train feature is identified. The images in the train feature that were having the same category as that of the test features were identified to be the similar images. The performance of the process is measured in-terms of performance metrics like Precision and recall.