Infrared Small Target Detection Based On Spatial-temporal Saliency

Human vision system(HVS)selectively deal with the information of human interest and ignore the important redundant information,which is the visual attention mechanism.Visual saliency simulates the visual attention mechanism of the human eye vision system.It can remove the redundant information in the image and quickly focus on the saliency area that may be caught by human eye.Therefore,HVS has been widely researched by the scholars and used in the target detection,image segmentation,video compression and so on.In this thesis,the infrared small target detection based on visual saliency is studied.To achieve the purpose of suppressing the background and enhance the target,the differences of feature of target and background is used in infrared small target detection.Infrared small target detection is mainly used in military field,which is an important part of infrared guidance system.How to design an efficient,robust and generalized infrared small target detection algorithm is an important research.The concrete research contents are as follows:(1)Four commonly used visual saliency models are studied.The frequency domain saliency model is chosen,which has high computational efficiency,simple principle and no prior information.Several representative models in the frequency domain are well studied.The phase spectrum of quaternion Fourier Transform is found,which can use more features of the image.Some suitable features to distinguish the target and the background are found by analyzing the feature of the target and background in the infrared image.(2)For the problem of target detection in infrared video,an infrared small target detection algorithm based on multi-feature is proposed,which includes strength feature,motion feature and direction feature.Simulation and testing of the algorithm are carried out and the result is analyzed.(3)For the problem of multi-target infrared image,an infrared small target detection algorithm based on image blocks is proposed.The candidate block is got by using a square window sliding in the image.The candidate blocks are judged by the method of central-surround difference and the suitable candidate blocks are detected with infrared small targets detection algorithm.Simulation and testing of the algorithm are carried out and the result is analyzed.(4)Matrix decomposition is studied in this thesis.The original three-dimensional structure in video is destroyed by matrix model.Based on this,an infrared small target detection algorithm based on tensor decomposition is proposed.The tensor model of infrared video is used and the objective function based on tensor is constructed.Tensor decomposition algorithm is studied and the sparse tensor is obtained as the final test result.Simulation and testing of the algorithm are carried out and the result is analyzed.