Researches And Applications Of Object Proposals Algorithm

When the human visual system processes a scene they will give priority to the visual attention to approximate objects in the scene and it makes the time of getting the scene shortest,these areas composed of possible objects called the object proposals.In the process of analyzing and handling images,the extraction skill in this region adequately simulates the character of seizing the features of target quickly in human vision and reduces effectively the detection time of background region and the corresponding calculation,as a result,it improves the accuracy rate of target detection and real-time.So focusing on the extraction skill in the object proposal is pretty significant to image processing and further development of computer vision field.Based on this idea,this article carries out the research work from the following several aspects.To begin with,aiming at the problem of producing a large number of background windows in the candidate region,then the secondary screening improved algorithm based on contour information is presented.The algorithm firstly applies the Prewitt operator to the Lab color space to extract image features,and then uses the object proposal method in the Binarized Normed Gradients algorithm to extract candidate regions,and finally uses contour information to determine the number of complete contours in the candidate window and the number of effective contours,and the specific amount to indicate the possibility of effective candidate region.Experiments are conducted on the PASCAL 2007 dataset,which achieves an average of 780 target candidate regions,and over 0.936 object recall on each image.When the target image or single background is relatively simple,the number of candidate boxes is less than 70.Experimental results show that the new algorithm can successfully reduce the number of object proposals based on the high recall rate.Secondly,according to the interference of the background and the change of illumination,we propose a new algorithm based on the candidate region.Firstly,using different visual effects process different illumination scenarios,eliminating this adverse factors of the illumination,then extracting the object proposal through the improved HSV color segmentation algorithm,removing the background interference,after contour detection on target candidate region is extracted,detecting the external contour of objects and elliptic check.In thesis,the algorithm and several commonly used circular mark detection algorithm was evaluated in German traffic sign database,the comparative experimental results show that.the algorithm has a good detection effect and can adapt to different illumination scene more.Finally,the paper uses the C++ programming language to build a test environment based on the object proposal circular mark detection algorithm,which provides the equipment support and experimental platform for further research.