V. Sheshathri ¹, S. Sukumaran ¹

Affiliations
1 Department of Computer Science, Erode Arts and Science College, Erode – 638112, Tamil Nadu, IN

Abstract

Objectives: The aim of segmentation process is to divide the image into homogeneous, self-consistent region or objects. The segmentation algorithms try to make systematic uses of some physically measured image features. Methods/Statistical Analysis: In the earlier research work, the researchers covered only multi resolution, quad tree structure, ant colony optimization and Otsu method for segmentation object in gray scale but not for color images. To overcome these issues, the segmentation for color image is focused on Enhanced Ant Colony Clustering (EACC) method. Findings: The proposed EACC method has three main parts. The first part is used to isolate the components of the given color image including RGB pixel values. The second part finds the clustering center with the help of combination of statistical and artificial selection. The last part implements EACC algorithm to segment on color image. In the existing method of ACO, the processing time takes longer time to segment the object. At the same time, while comparing the threshold value on existing method is lower than current proposed method of EACC. Applications/Improvements: In this research work, the proposed method achieves better segmentation in color image for the data sets Oxford Flowers 17, Weizmann Horse and MSRC dataset. Further, this work may be extended to different type of images such as, multiband or multispectral images, satellite images, etc. Finally, the experimental results are shown through Mat Lab R2013a.

Keywords

Enhanced Ant Colony Clustering, RGB Pixel Values, Segmentation, Self-Consistent Region, Threshold Value.

Full Text

   

S. Janarthanam ¹, S. Sukumaran ², M. Shanthakumar ³

Affiliations
1 Department of Computer Science, Gobi Arts and Science College, Erode – 638452, Tamil Nadu, IN
2 Department of Computer Science, Erode Arts and Science College, Erode – 638112, Tamil Nadu, IN
3 Department of Computer Science, Kamban College of Arts and Science, Coimbatore, IN

Abstract

Objectives: The objective of the examined active salient component classifier is to extract the image features from the visual substance through a novel non-parametric example based positioning methodology to utilizing comparability and learning strategies. Methods/Statistical Analysis: To establish the authenticity to avoid improper retrieval on online based image classification system called, lite Dynamic Pattern Extraction Algorithm (DPEA) for Image Classification and salient image extraction algorithm is designed. DPEA Algorithm on each benchmark dataset works on user query basis and reducing the overhead incurred during user query processing by applying labeling process. Next, by retrieving salient features on images with zero mean for each client query and each image retrieval reduces the execution time and complexity as the database does not maintain the related features. Finally, the Potential image Classification and retrieval prevents the unauthorized user modification on image data, therefore improving the reality. Here a Web Image Dataset (NUS-WIDE) created by lab of media search in National University of Singapore. The computer vision CIFAR-10 dataset, Modified National Institute of Standards and Technology database of USA (MNIST) dataset is used for experiment. A series of simulation results are performed to test the image retrieval efficiency, execution time, performance fitness for obtaining efficiency of image data handling and measure the effectiveness of DPEA Algorithm. Findings: The tests were directed on the dataset for image classification and retrieval on each region of the images. The Proposed DPEA offers better performance with an improvement of the data confidentiality by 18.72%, reduces execution complexity by 7.98%, reduces average retrieval overhead by 6.65% and also minimize retrieval complexity by 15.57% compared to existing algorithms of LFDA, PCA and PCCA respectively. Application/Improvements: It can be further extended with implementation of new retrieval techniques with different parameters which improves more retrieval efficiency and performance integrity.

Keywords

Bit Streams, Component Analysis, Dynamic Patterns, Feature Extraction, Image Retrieval, Salient Features.

Full Text