Research Of Visible Object Detection Based On Sparse Representation

With the development of information technology, people gradually pay more atten-tion to object detection and recognition based intelligent image analysis and processingtechnology. In this paper, we concern about the object detection issue based on Bag ofWords model. We research on a general object detection algorithm for natural imagesbased on Bag of Words model and an object detection algorithm for remote sensing seaimages based on sparse codebook.When facing the object detection issue coming with natural images, we design ageneral object detection algorithm frame based on Bag of Words model. In respect offeature extraction, we confirm the sparse feature to describe the images after a number ofanalysis and comparison. Then redundancy information in images is reduced. In respectof constructing object description model with codebook, we add our semantic neighbourconstraint coding model on traditional histogram model for enhancing description capa-bility on rigid objects. In respect of object localization, we combine linear SVM (SupportVector Machines) classifier with Branch and Bound algorithm for search acceleration.Experiments indicate that the improved model can provide better description on rigidobjects than traditional model. And the detection algorithm performs much faster thansimple sliding window search.When facing the object detection issue coming with remote sensing sea images, wepropose an object detection algorithm frame based on sparse codebook. When concern-ing about the abundant texture coming with high resolution remote sensing images, wepick local patches as image features. We construct our sparse codebook based on the lo-cal patches and Bag of Words idea. And the codebook is sparse for sea area. Then webuild histogram with neighbour patches, and apply linear SVM classifier to classify lo-cal patches. Finally, the object is localized after sea area and object area are segmented.Experiments indicate that the sparse codebook describes high resolution remote sensingsea images excellently. And the object detection algorithm for remote sensing sea imagesbased on sparse codebook can perform high relevance ratio with low false alarm rate.