Search

Abstract: The broad region outside the classical receptive field (CRF) of a vision neuron, known as the non-classical receptive field (nCRF), exerts a robust modulatory effect on the responses to visual stimuli presented within the CRF, and plays an important role in visual information processing. One possible role for the nCRF is the extract object contours from disorderly background textures. In this study, a multi-scale integration based contour extraction model, inspired by the inhibitory and disinhibitory interactions between the CRF and the nCRF is presented. Unlike previous models, our model not only includes both the simple and complex cell mechanisms but also introduces pre-processing of the external information by the retinal ganglion cells at an early stage. The multi-scale representation of a physical scene acquired through such pre-processing was filtered through Gabor filters, and then inhibited or disinhibited at different spatial locations on different scales until a final response was obtained. Our results show that by introducing this kind of mechanism into the model, numbers of non-meaningful texture elements can be removed significantly, while at the same time, the object contours can be detected effectively. In addition to the superior contour detection performance in comparison to other contour detection models, our model provides a better understanding of the role of the nCRF and a novel approach for computer vision and pattern recognition. [Copyright &y& Elsevier]

Abstract: Recently hashing has become attractive in large-scale visual search, owing to its theoretical guarantee and practical success. However, most of the state-of-the-art hashing methods can only employ a single feature type to learn hashing functions. Related research on image search, clustering, and other domains has proved the advantages of fusing multiple features. In this paper we propose a novel multiple feature kernel hashing framework, where hashing functions are learned to preserve certain similarities with linearly combined multiple kernels corresponding to different features. The framework is not only compatible with general types of data and diverse types of similarities indicated by different visual features, but also general for both supervised and unsupervised scenarios. We present efficient alternating optimization algorithms to learn both the hashing functions and the optimal kernel combination. Experimental results on three large-scale benchmarks CIFAR-10, NUS-WIDE and a-TRECVID show that the proposed approach can achieve superior accuracy and efficiency over state-of-the-art methods. [Copyright &y& Elsevier]