ABSTRACT

A major area of current research in data mining is the field of medical diagnosis. In the present study using the Breast cancer Wisconsin data sets, a feature selection algorithm Modified Correlation Rough Set Feature Selection (MCRSFS) predicts both diagnosis and prognosis by comparing several data mining classification algorithms. In the proposed approach, in level 1 of feature selection, features are selected based on rough set with different starting values of reduct. In level 2 features are selected from the reduced set based on the Correlation Feature Selection (CFS). Experiments show the proposed method is effective by comparing with others in terms of number of selected features and classification performance.

Digital image watermarking is implemented using fuzzy logic approach based on transform domain. Fuzzy logic is an approach based on "degrees of truth" rather than usual "true or false" i.e., Boolean logic. In this paper, we use a reference image instead of original image. To get a reference image Directive contrast and Threshold value are calculated for middle and high frequency bands. A transform domain with the help of fuzzy logic system is used to embed the watermark in the reference image. Here fuzzy logic is used to find the alpha value that is used to embed the watermark. Watermark embedding is based on DWT and Singular Value Decomposition (SVD). After decomposing the cover image into four bands, SVD is applied to one of the frequency bands, and embed the same watermark data by modifying the singular values. Attacks are also performed on watermarked image. Quantitative measures like PSNR, NCC are also calculated to test the watermarked image and also extracted watermarks with and without attacks General Terms Watermarking, Security, Algorithms.

In this work, a watermarking algorithm is proposed using Genetic algorithm and discrete wavelet transformation. The algorithm proposed is to improve both robustness and fidelity of the watermarked image. Fuzzy Inference system is used to determine the embedding strength based on HVS properties of the image. To test the robustness different attacks are performed and the NC (Normalized cross correlation) is computed. The algorithms showed better results and good quality for watermarked image.

Edge detection is one of the most commonly used operations in image processing and pattern recognition. Edge detecting in an image significantly reduces the amount of data and filters out useless information, while preserving the important structural properties in an image. In this paper, edge detection methods such as Sobel, Prewitt, Robert, Canny, Laplacian of Gaussian (LOG), Expectation-Maximization (EM) algorithm, OSTU and Genetic algorithms are also used for segmenting. A new edge detection technique is proposed which detects the sharp and accurate edges that are not possible with the existing techniques. This implemented edge detection technique will be improved by combining it with other types of filters namely Weiner, STD, Hormonic, Geometric filters to remove the noise from the image. The proposed method is applied over large database of color images both synthetic and real life images and performance of the algorithm is evident from the results with different threshold values for given input image which ranges between 0 and 1. When the threshold value is 0.68 it is noticed that the sharp and accurate edges are detected.

No abstract

Feature selection aims to select subset of original features. It removes unrelated, redundant or noisy data from the problem domain. Rough set theory is often applied to feature reduction using the data alone, requiring no additional information and widely used for classification tool in data mining. Clustering, a form of data grouping, groups a set of data such that the intra-cluster similarity is maximized and the inter-cluster similarity is minimized. In this paper, k-means clustering algorithm is applied to partition the given information system and further rough set theory implemented on the data set to generate feature subset. The classification process by means of SVM is performed by using the remaining features. Wisconsin Breast Cancer datasets derived from UCI machine learning database are used for the purpose of testing the proposed hybrid model and the success rate of hybrid model is determined as 99%. General TermsPattern Recognition, Machine learning.

No abstract