| Summary: | Research in image processing has become an important area due to the remarkable growth in digital images and widespread of image processing applications in various domains. Ensemble methods are learning algorithm that construct a set of classifiers and then classify new data points by taking a weighted vote of their predictions. Ensemble method used in this thesis are random forest, bagging and multi class classifier. From previous studies, one of the problems in this research area was a large number of categories for single object images (SOI). This results in difficulties to find the best techniques to produce a framework with higher accuracy approach. The second problem was detection of the best feature among all features, of which feature will give a better accuracy. The third problem identified the best classification method in order to produce high accuracy result. This research proposed a classification framework for single object images in multiple categories using ensemble method. Image processing using single object images (SOI) has become one of the research fields which involved determining the classification of the main object in the image. The proposed framework comprises several image processing methods that are crucial in assuring the accuracy of classification; (i) Image segmentation, (ii) Object identification, (iii) Feature extraction, (iv) Feature Selection, (v) Image classification. Each mentioned method was already tested and compared in order to find the best combinations of procedures to produce the highest accuracy framework. The proposed framework has been trained, tested and compared using three (3) dataset of image collections; (i) Amazon images, (ii) Google images and (iii) Caltech-101 images. Amazon and Google images dataset represent SOI collection while Caltech-101 image represents the collection of multiple object images (MOI). These datasets were used for comparison processes in order to prove the ability of the proposed framework to process various types of images. MATLAB is used for the image processing task, whereas WEKA is used for the classification task. From the experiment, the Amazon dataset gives 99.00% accuracy while Google provides 92.65%. Meanwhile, Caltech-101 produces lower accuracy, 85.00%. In conclusion, the result prove that the classification model based on ensemble classifiers produces the highest accuracy when using SOI dataset compared to MOI.
|
|---|