Construction And Application Of Image Screening Model For Ancient Villages Based On SIFT And BP Neural Network

Ancient villages are the crystallization of Chinese farming culture for thousands of years,and have a long history and rich cultural value.Building digital archives for the ancient village is a means of protection of ancient villages.But the wide distribution of so many ancient villages make it difficult to collect digital archives.To solve this issue,crowdsourcing is applied to the acquisition of digital resources,especially images of these ancient villages.Although crowdsourcing can quickly and widely acquire to these images,but there are still some problems.Due to different skills and devices,some ancient village images taken by participants in crowdsourcing are distorted.In addition,a participant may take multiple images on the same subject with similar angle continuously,and multiple participants may take images on these objects with similar angle and scale.As a result,some ancient village images are near-duplicate.In order to ensure the quality of the ancient village digital archives,it is necessary to screen the images delivered by the participants efficiently and objectively.In this paper,we firstly study and summarize the existing image feature extraction method and image quality evaluation method.Then,we proposes a screening model for ancient village images based on SIFT and Back Propagation neural networks to screening ancient village images.The model has two stages: screening of near-duplicate images based on SIFT and block color histogram,image quality screening based on Back Propagation neural networks.In screening of near-duplicate images based on SIFT and block color histogram,we firstly extract the image SIFT features to match images.Then,we calculate and compare the block color histogram of matched image pairs to judge whether they are near-duplicate.We compute the NIQE score of these near-duplicate images,keep the best and screen off others.In image quality screening based on BP neural network,we first calculate the image ambiguity,noise estimation and NIQE quality score as features.The BP neural network is constructed by using these values as well as some image attributes.The operator of an image is divided into two categories: keeping or screening off.We use the manually labeled data set to train the neural network.For an image to be screened,we just need to calculate the ambiguity,noise estimation and NIQE quality score,input these features into the convergence of the neural network,then we obtain the result of image quality screening.Finally,we apply the proposed model to the ancient village cloud service platform,use the ancient village image test set to test the model and verify the availability of the model.Compare the results of the proposed method with the artificial screening to verify the accuracy.For the image quality screening based on BP neural network,the logistic regression and the support vector machine are used for the same training set and test set respectively.The experimental results show that our model outperforms others.