Research On Image Super Resolution And Object Recognition Based On Object Model

The massive popularity of portable shooting equipment,massive video image information and provides abundant data resources for us,the video image plays an important role in the field of public safety,production and life field,military field.But because the mass of the image information,the traditional method of eye recognition efficiency is low,unable to meet the basic needs,so it is necessary to find an image recognition algorithm to help interested in looking for our goals and to deal with.Due to the limitation of the portable equipment and shooting shooting environment,shooting angle and shooting level,usually bad image quality,fuzzy objective information,to a certain extent for target recognition,so the research on the target recognition algorithm based on super resolution image has important theoretical significance and application value.This paper from the content and structure of the image of the restored image details using the image prior model,super-resolution image reconstruction,while the transmission,based on coherent detection of frequency offset estimation algorithm,and then use the reconstructed super-resolution image target recognition algorithm research,effectively improve the accuracy and efficiency of target recognition.The main research work is as follows:First of all,in the process of image recognition for quality factors(such as noise,fuzzy and down sampling)caused the problem of low recognition rate,proposed a super-resolution reconstruction method for non local anisotropic prior model based on the image effect is more clear.Estimation of quality factor of the fuzzy degree,reducing noise level of the source image,combined with the natural image edge information and texture information,and consider the repeated structure information of the image inside,using the structure of non neighborhood regions in the image similarity,provide complementary information of the image,and use this information to estimate the pixel distribution model,considering the to the specific type of image level in the vertical direction,the weight model of anisotropic respectively to calculate the weight,so as to achieve the purpose of image detail estimation,provide the guarantee for the image target recognition algorithm with high accuracy.Remove the influence of noise,fuzzy and down sampling quality factors in the image effectively,with more abundant information display,which can grasp the effective image feature feature extraction and training process,the establishment of effective image recognition algorithm.Experimental results show that the algorithm can effectively remove image noise and blur,improve image resolution,and preserve the original gradient features of images.At the same time,in order to better transmit high quality image signal,in optical communication,especially in the coherent detection of frequency offset estimation algorithm research,32-QAM modulation format is proposed for estimation algorithm a work in feedforward mode and does not require additional data aided frequency.We propose a FFT method that extracts and amplifies the QPSK inner loop of 32-QAM.This method keeps the fast estimation speed of the traditional FFT method,and keeps the large frequency offset estimation range of the baud rate /8.Second,an asymmetric inverse layout target representation model(NAOM)is proposed to represent the shape change of target object in order to detect the local spatial relationship between the target and the whole.The target detection model is established in order to identify the optimal way to effectively detect targets;we will target detection as an inverse distribution process,which focuses on learning target feature description and NAOM target model proposed using the grid edge orientation histogram(G-HED)features the local information of the target object to better describe at the same time,through the study of edge feature descriptors,target identification and model learning algorithm,the target classifier,especially the NAOM model and the G-HED descriptor and effectively describe the target should be global and local changes,characteristics of the target has a stronger and more accurate expression ability.Third,through the algorithm of super resolution reconstruction of image and space by NAOM object representation model inspired,proposed a target recognition algorithm based on NAOM model,improves the recognition accuracy and speed efficiency,understanding for the subsequent image provide effective help.The target detection algorithm of NAOM model is a three layer structure based on the proposed Hough detection algorithm of weighted voting,quantized local block classifier detection value,effectively improve the efficiency of the identification and detection of target block.Analysis and experimental results have proved the superiority of the research topic of this article in the algorithm,especially with other similar target detection algorithm,which greatly enhance the target recognition and detection algorithm efficiency and accuracy.