Abstract
Kernel classifiers based on the hand-crafted image descriptors proposed in the literature have achieved state-of-the-art results in several dataset and been widely used in image classification systems. Due to the high intra-class and inter-class variety of image categories, no single descriptor could be optimal in all situations. Combining multiple descriptors for a given task is a way to improve the accuracy of the image classification systems. In this paper, we propose a filter framework “Learning to Align the Kernel to its Ideal Form(LAKIF)” to automatically learn the optimal linear combination of multiple kernels. Given the image dataset and the kernels computed on the image descriptors, the optimal kernel weight is learned before the classification. Our method effectively learns the kernel weights by aligning the kernels to their ideal forms, leading to quadratic programming solution. The method takes into account the variation of kernel matrix and imbalanced dataset, which are common in real world image categorization tasks. Experimental results on Graz-01 and Caltech-101 image databases show the effectiveness and robustness of our method.
Highlights
Classifying multiple object categories is a challenging task for computer vision systems
Because the image features that proposed in the literature are variable, and kernel value of image pairs are much different, are more influenced by the variance of the kernel matrix combined; second, the training image dataset is often imbalanced, and most of “filter” methods suffer from imbalanced training dataset
Because LAKIF and Kernel Target Alignment (KTA) methods are filter methods, we report the training time of kernel weight learning and SVM training seperately
Summary
Classifying multiple object categories is a challenging task for computer vision systems. Learning the weight of kernel combination has been explored in the literature, roughly categorized into “filter” based or “wrapper” based. Because the image features that proposed in the literature are variable, and kernel value of image pairs are much different, are more influenced by the variance of the kernel matrix combined; second, the training image dataset is often imbalanced, and most of “filter” methods suffer from imbalanced training dataset. We propose Learning to Align the Kernel to its Ideal Form(LAKIF) framework for learning the optimal combination of kernels. It is a “filter” framework, and our aim is achieving comparable performance with “wrapper” methods while keeping the training efficiency. The framework handles imbalanced training data, which is common in object recognition tasks
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.