An Efficient Indexing Approach for Content based Image Retrieval

In this paper an efficient indexing and retrieval technique is proposed for identification of plant images. The plant images have been retrieved as herbs shrubs and trees based on color and texture features. Plant images have rich content of information such as leaves, bark and stem etc. This content can be extracted as various content features. Plant image database often represent the image objects as high-dimensional feature vectors and access them via the feature vectors and similarity measures. Hence in order to retrieve plant images an indexing method is proposed based image retrieval to improve the retrieval performance. In this work color and texture features are extracted for characterizing leafy mass and stem images of plants. There are 20 images of five plant species, amounting to total of 100 images, herbs, shrubs, and trees. Hence, totally there are 300 images of leafy mass and bark of 15 plant species. All the images are captured in natural environment using a digital camera. In the pre-processing stage, the images of leafy mass and bark with complex background information and noise is filtered out by using Unsharp filtering method. In the segmentation method 2D-OTSU threshold based segmentation technique is used to separate the object from the background For Ex: small leaves surrounded by bright sky, small openings among dense canopy, and mixed pixels. For feature extraction the Modified Color Co-occurrence Matrix (MCCM) and Gabour Filter is used to retrieve the color and texture of the images from the database. Usually the retrieval performance of an image retrieval system is greatly influenced by different similarities or distance measures. We have used Euclidean distance measures for similarity matching. Finally for retrieval of the images from the database Fast Indexing scheme for CBIR method is used. Then multi-dimensional indexing technique is employed using tree structures, R* tree. The effective indexing and fast searching of images on bases of visual features pose a significant issue in CBIR. We have used R*-Tree structure to achieve better performance and efficiency. A graph is plotted to compare the retrieval result of both technique. Indexing Technique gives better retrieval rate.