الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 117 |  |  |
| | | from | 117 | | |
Abstract
Designing object recognition systems that work in the real world is challenging task due to various factors including occlusion, clutter, articulation and noise that make the extraction of reliable features quiet difficult. Furthermore, feature useful to the recognition of one kind of object may not be effective in the recognition of another kind of object. Thus, the recognition system often needs thorough overhaul when applied to other types of images different from the one for which the system is designed. This is very uneconomical and requires highly trained experts. The purpose of incorporating learning into the system design is to avoid the time consuming process of feature generation and selection and lower the cost of building object recognition systems. Evolutionary computation provides a systemic way of synthesis and analysis of object recognition system. With learning incorporated, the resulting recognition system will be able to automatically generate new features on the fly and cleverly select a good subset of features according to the type of objects to which they are applied. The system will be flexible and can be applied to a variety of objects. In our work, feature is composed by Genetic Algorithm (GA), where each feature has an image patch and an appropriate sequence of images transformations that are selected by the GA. The discrimination ability of a feature is represented by a classifier. Classifier fusion (Ensemble) method is used to combine the output of the multiple base classifiers’ outputs and build a final classification model. The performance of machine learning method (classifier) is heavily dependent on the choice of data representation (or features) on which they are applied. When a classifier does not scale well to high-dimensional data, this problem is called the curse of dimensionality. To overcome this problem of dimensionality, a good solution can often be made by changing the algorithm, or by pre-processing data into a lower-dimensional form. We investigate several base classifiers’ algorithms such as Artificial Neural Network(ANN), k-Nearest Neighbor (kNN), and linear/non-linear Support Vector Machine (SVM). Moreover, different image descriptors such as Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP) and pooling operation are used to represent the features in a robust form. Two alternative ensemble algorithms (AdaBoost and Random forest) are employed to aggregate the prediction of all features. The proposed framework try several combinations of base classifier, image descriptor and ensemble type to find the best combination that obtains the maximum accuracy for a given dataset. Furthermore, we implement our proposed framework to process classifiers concurrently to enhance the running time. Experimental results showed that the proposed method has a competitive performance over contemporary methods for different object datasets.