الفهرس 
Feature-based Object Recognition ApproachesOnly 14 pages are availabe for public view
 |  | from | 119 |  |  |
| | | from | 119 | | |
Abstract
This thesis introduces a new object recognition framework which includes various object recognition approaches based on PCA, LDA, and the K-NN classifier. This classifier is used with different distance criterion which are Euclidean distance(ED), Modified weighted Euclidean distance (MWED), Manhattan distance (MD), Modified weighted Manhattan distance (MWMD), and cosine distance (COSD) to determine the best distance metric. A color image vector representation model is introduced in this framework to represent each image instead of using separated color components as found in the literature. We also use a virtual samples technique which is based on rotation, mirror, and image contrast adjustment to generate virtual samples (new input samples) to improve the recognition rate of OR. We further introduce an OR approach based on Majority Voting Process (MVP). Due to the importance of color information in OR, the framework also includes a variety of color spaces (RGB, IRGB, YIQ, YCbCr, YQCr, CMY, CMYK, HIS, HSV, L*a*b*, YUV, and CcMmYK). Finally, a high-performance object recognition system is concluded from the framework based on some experimental results which show that the usage of vector representation model, virtual samples, CcMmYK color space, PCA, and K-NN introduce great improvements in the recognition rate of the OR system. To support our claim, some experiments are introduced using COIL-100 and ALOI color object databases. The results of the experiments are compared with [18], [38], [57], and [60] on Coil-100 database to show the effectiveness of the proposed work that ended up with improvements in the recognition rate by approximately 21.7%, 2.3%, 8.42%, and 2.36%. We also compare with [38] on ALOI database that ended up with improvements in the recognition rate by approximately 15.45%.