Digital Image Processing

V. Subbaroyan ¹, S. Karthik ²

Affiliations
1 Sathyabama University, IN
2 Cognizant Technology Solutions, Chennai, IN

Abstract

In this paper an attempt has been made to compare three different approaches of Object Recognition namely, Gradient based, Histogram based and Texture based methods. For a realistic approach common household articles with uniform colour properties have been taken up for this study, instead of standard images. An evaluation of the comparative study has been made and the results have been tabulated. We believe that this study will be useful in choosing the appropriate approach in object recognition for service robots. In this paper, we evaluate an object recognition system building on three types of method, Gradient based method, Histogram based method and Texture based method. These methods are suitable for objects of uniform color properties such as cups, cutlery, fruits etc. The system has a significant potential both in terms of service robot and programming by demonstration tasks. This paper outlines the three object recognition system with comparison, and shows the results of experimental object recognition using the three methods.

Keywords

Correlation, Gradient, Histogram, Texture.

Full Text

     

S. Karthik ¹, H. R. Mamatha ², K. Srikanta Murthy ³

Affiliations
1 Department of IS&E, PES School of Engineering, Bangalore, IN
2 Department of IS&E, PES Institute of Technology, Bangalore, IN
3 Department of CS&E, PES School of Engineering, Bangalore, IN

Abstract

Optical Character Recognition (OCR) is one of the important field in image processing and pattern recognition domain. Many practical applications uses OCR with high accuracy. The accuracy of the Optical Character Recognition system depends on the quality of the features extracted and the effectiveness of the classifier. Here we are proposing a novel method to recognize the printed kannada vowels. Kannada script has large number of characters having similar shapes and also the complexity is font dependent, which means the same characters in a class, may vary in structure for different fonts. Hence a method, which makes use of image zoning and the Run Length Count techniques to extract the features have been proposed. The methodology uses Naive Bayes classifier, K-Nearest Neighbor classifier for classification. The method experimented on a dataset, which consists of samples from 69 different fonts, and a maximum of 97.44% recognition accuracy is achieved.

Keywords

Optical Character Recognition, Naive Bayes Classifier, K-Nearest Neighbor Classifier, Zoning, Run Length Count.

Full Text

     

K. Vasanth ¹, S. Karthik ²

Affiliations
1 Department of E.E.E, Sathyabama University, Jeppiaar Nagar, Rajiv Gandhi Salai, Chennai-119, IN
2 Cognizant Technologies Solutions, Chennai, Tamilnadu, IN

Abstract

In this paper, a fixed 3×3 window size switched algorithm for removal of the image noises such as fixed impulse, random impulse noise, zero mean Gaussian noise, mixed noise and artifacts such as blotches, stripes, streaks for gray scale images for increasing noise densities is presented. The proposed method uses a tree based method to classify the noisy pixels and noisy free pixels. Being a switched algorithm the median is calculated only after identifying the pixel is noisy. At higher noisy environments the corrupted pixel is replaced either by the median or by mean of non noisy pixel in the current window or the preprocessed neighborhood pixel based on the classification. The quantitative and qualitative performance of the proposed algorithm outclasses many of the standard filters in terms of different artifact removal with edge preservation. The proposed algorithm found to give better results in the removal of fixed impulse noise up to 80%, and preserves edges up to 70%, zero mean 0.9% variance Gaussian noise. The proposed algorithm even found to suppress mixed noise of 10% impulse noise plus zero mean 0.3% variance Gaussian noise composition.

Keywords

Switched Median Filter, Modified Decomposition Filter, Threshold Decomposition Filter.

Full Text