1. UKM Learning and Research Repository
2. Master Thesis / Tesis Master
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	An enhanced method of multi unseen objects class recognition using saliency based visual attention
Authors:	Vahid Alizadeh Sahzabi (P53641)
Supervisor:	Khairuddin Omar, Prof. Dr.
Keywords:	Universiti Kebangsaan Malaysia -- Dissertations Dissertations, Academic -- Malaysia Computer vision Pattern recognition systems Image processing -- Digital techniques
Issue Date:	3-Sep-2012
Description:	In pattern recognition and machine vision area of research a robust method to categorize new objects is always on demand. Therefore, this study aims to propose a method for object class recognition in images with less knowledge and fast. The method is more concentrate on unseen object and tries to classify them at a glance as main problem. The objective of the research is to classify all objects which are existing in all images based on Attention Based system region selection and using the fuzzy logic for final decision making. Hence, it needs to extract some discriminative features from the images; furthermore the problem of current research is when unseen objects appear in the image. This method is consisting of five steps; at the first step, the algorithm starts with region selection. Then in the second step, it recognizes each object that has been existed in the database. In third step, if all objects had existed in database so all objects have been classified successfully, otherwise proposed algorithm starts to classify unknown objects based on neighbour objects. In the fourth step, the decision making to classify unseen objects should be based on some parameters such as, firstly matching keypoints extracted from SURF algorithm, secondly find shape matching between similar objects in shape and colour and thirdly the objects which are mostly detected. Finally in last step, fuzzy measure will be used to classify unknown object or if not classified so, it is a new category. It is precisely like human being manner. In the other words, similarity between objects in images will increase the likelihood of correct classification of unseen objects. The structure of the proposed method has a big different with the other common methods; in this algorithm object will be classified at first glance and Speed Up Robust Features (SURF) algorithm has been used to detect and recognize objects. The algorithm is tested in both on standard CVIU image dataset and fifty six new images. Obtained results have shown that proposed algorithm is more adaptable in case of classifying unseen objects in compare with previous methods. The accuracy has improved 4% of pen detection in compare with Winn's method. Also SURF has good accuracy in time and error rate -1000 times faster and average 12% reduces error rate- in compare with SIFT and PCA-SIFT.,Certification of Master's/ Doctorial Thesis" is not available
Pages:	103
Call Number:	TA1634.S237 2012 3 tesis
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.