Object classification through selsected image segments

• Main Author: Teng, Serene Shu Hui.
• Other Authors: Teoh Eam Khwang
• Format: Final Year Project
• Language: English
• Published: 2011
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Computer hardware, software and systems
• Online Access: http://hdl.handle.net/10356/46211
• Institution: Nanyang Technological University

Object and face data processing has been an active research fields for many decades due to its various application in different areas such as security systems, video surveillance applications, biometric systems and information security. However, in object classification only a few works have been reported in implementing in a real-time traffic surveillance system. In particular, family car classification problem is not investigated. The main focus of this project is to develop suitable approaches for evaluation of the contribution of image segments to object classification. Firstly, large amount of dataset of raw images of family of objects need to be collected. Segmentation of background is being done so as to reject as much ‗non-car‘ of the image as possible. The data sets will consist of car family with background and without background. Usually, car in the images might not be standard; images are in different pose n position. In order to generalize the car positions of the data sets, cropping and alignment process have to be done as a pre-processing. The comprehensive objective of this project is to focus on classification of car family members from a non-family member of car using Continuous Fusion rule algorithm. Feature of the car images from the car family are first being extracted out by the dense Scale Invariant Feature Transform and Gabor features. Then the Modest Adaptive Boosting classifier is used to train the extracted feature. The classifier will select and remove the redundant features after the training. In addition, algorithms are compared in terms of output class of discrete and continuous classifiers. Finally, the final results are obtained by fusing of the output from the classifier using object segments classes that leads to superiority of the employment of the proposed algorithm. The classification performance is evaluated by three types of error, the False Negative error Rate, False Positive error Rate and Total Error.