Image Set Enhancement And Management Based On Composition Adjustment And Similarity Analysis

In recent years,with the vigorous development of computer vision science,its application technology has had a profound impact on all aspects of people's lives,in-cluding camera beauty,face recognition,unmanned driving and so on.Image process-ing is the foundation and core of computer vision.Both computer vision and image processing are inseparable and complementary.The research work of this paper be-longs to both the computer vision and the image processing field.A method of image set enhancement and management based on the composition adjustment and similarity analysis is designed and studied.On the one hand,the traditional image set enhance-ment method is generally to adjust the brightness of the image,denoising and sharpen the image,by contrast,we enhance the image set by adjusting the composition of the image.On the other hand,the management of image sets is generally classified ac-cording to the time or location of the image,by contrast,we put forward a method to classify the image set according to the similarity of human visual characteristics between images.First,we propose seam carving based and cropping based composition adjustment methods.Traditional seam carving technology can scale the image and will not cause distortion in content scale.Based on this feature,combined with the detection of salient region,we achieve the goal of image composition adjustment by moving the salient region to a more reasonable place.The energy map is initialized and updated according to the location of the salient region,which effectively solves the phenomenon that the foreground of the adjusted results is easy to distort.The clustering phenomenon of the seams is effectively solved by constantly updating the energy map before searching for the seams.The efficiency of the algorithm is greatly improved by traversing only part of the energy map while each time the dynamic programming algorithm is executed.In the study of cropping based composition adjustment,the objective equation is set up based on a series of composition rules,and the result of our composition adjustment is obtained by optimizing the objective equation to find the best cropping window.Due to the use of "downsampling-calculation-upsampling" approach,the execution speed of our algorithm is not affected by the resolution of the input image.Second,we propose an image similarity analysis method based on human visual characteristics.Firstly,according to the color information of the image,the similar-ity analysis is carried out.We analyze the foreground and background similarity of the image respectively,and then synthesize it according to their respective area.The fingerprint of the image is calculated and the similarity between the images is judged according to the correlation of the fingerprints among the images.Then,the similarity analysis is made according to the content information of the image.The content map of the image is obtained by using the threshold segmentation method based on Otsu,and the joint gradient domain between the images is calculated to get the content influ-ence factor between the corresponding pixels,thereby the content similarity between the images is calculated.After that,we analyze the similarity of two images based on both the color and content information of the images,and the result is more accurate through our fine-to-coarse operation.Finally,the above three methods are combined,and the final image similarity is refined according to the time and GPS information of the image.According to the user study,our result error rate is 0.098.Last,we propose a method of image set enhancement and management based on composition adjustment and similarity analysis.The image set enhancement is realized by using the composition adjustment.The data set is expanded by two times,which not only increases the user's selection,but also enhances the similarity correlation between the images in the image set,which provides the basis for the subsequent image classi-fication algorithm.Afterwards,the similarity analysis method is used to calculate the similarity value between any two images in the image set,and the correlation graph is constructed.The graph segmentation operation is performed on the correlation graph to realize the classification management function of the image set.