An enhanced approach for morphological shape representation and image retrieval

Abstract

Our work deals with a new Shape representation and Image Retrieval technique using morphological approach for binary images In many shape analysis problems only binary image is needed where object is usually depicted with value ones and background is depicted with value zeros In such cases binary image mathematical morphology is enough While selecting the features it can be grouped into the following types 1 colour representation 2 texture representation 3 local features and 4 shape representation In this proposed work the feature shape is selected using morphological skeleton representation In this Shape representation technique the shape can be represented by integrating the improved octagonal tracing algorithm and disk selecting algorithm In the past representation techniques octagons of same size are used throughout the shape and also it does not exhibit the disk selection order which makes the technique so complex and omission of smaller parts The modified octagonal tracing algorithm makes the shape as combination of various octagons from higher size to smaller size so that smaller parts are filled and it tells the order directly and thereby recursion is not happening in disk selection process In this Image Retrieval technique we use special parameters for angle or rotation of an object even if an image is tilted so that matching is perfect in our technique The matching parameters are taken in 12 directions and in effect efficiency of retrieval in our technique is higher than the efficiency of previously adapted techniques Our methodology includes Data Acquisition Pre processing Feature selection Feature extraction Restoration and Image Retrieval using similarity measures To validate our method the performance measures like precision recall are done Also the evaluation methods namely error rate retrieval efficiency are calculated From the set of database images we calculated the value of mr mn mb and M with the query images From the obtained values the images whose similarity measures values are closely matched with the query images are retrieved successfully Our algorithm handles shapes having rotational and scale changing features effectively We performed retrieval experiments on Kimia data sets and MPEG 7 data sets Our results are comparable to the best results available In this research work investigation is carried out to find an efficient method of developing a computationally efficient shape retrieval system for morphology based shape representation schemes while reducing the generation of noise after decomposition with minimum representation points is created During the noise removal the OMSD algorithm operates efficiently on a noise free environment and low noise environment The sensitivity offered by the algorithm is not sufficient for some image processing applications In order to overcome the sensitivity a relaxation is offered while finding the repeated erosions Such operations are called soft operations Since soft erosion is more immune to noise the required noise immunity for the OMSD algorithm is acquired The procedure for the algorithm is the same as that of the mathematical morphology based OMSD but the morphological operations are implemented using soft morphology Our proposed method is validated by performing a large set of evaluation tests on two datasets of shapes It also proves that the shape features taken from the shape representation algorithms namely octagonal tracing algorithm and disk selection algorithm and the similarity measures in image retrieval analysis using both the datasets namely Kimia dataset and MPEG 7 dataset It provides better discrimination compared to that based on four different existing techniques namely Inner distance Image motion Shape matching and Shape decomposition which are well known approaches used in image retrieval application The results of the experiments show that our method is an effective economical and flexible approach for shape representation and retrieval and further illustrate that this method is highly consistent and reproducible Among the various approaches to shape representation and retrieval the method based on morphological feature descriptors attracts more of our attention because using the information from the region to describe a shape is the most straightforward idea Using the shape primitives to represent the skeleton points has the advantage that when image is processed Also the processing procedures for solving the problem of invariant under scale translation and rotation are reasonably simple