Research On Image Scaling Distortion Suppression Technology And Classification Algorithm

The 21 st century is an era of information explosion.Compared with sound and text,images can enable people to obtain information mor e intuitively.It contains a larger amount of information,more flexible receiving methods,and more direct transmission methods.People need to scale,observe and classify the acquired images,and extract useful information from the images,so as to improve production efficiency and bring convenience to life.If you want to obtain detailed information of the picture through long-range photos or monitors,you need to scale the obtained image,but after the image is scaled,there will be block distortion or "mosaic",which will affect the image quality.At this time,image classification accuracy will decrease.This paper studies a variety of schemes that can effectively suppress image distortion during the scaling process,including nearest neighbor interpol ation,bilinear interpolation,elastic model,edge detection interpolation,bicubic interpolation,etc.,and carry out a comparative test on the above methods.Aiming at the problem that the bicubic interpolation algorithm has a large amount of calculations,this article divides the image into flat and complex regions by calculating the variance of the image,and performs bilinear interpolation and bicubic interpolation switching modes on the image for different regions.Compared with a single method,the improved bicubic interpolation algorithm can obtain a more effective reduction of image scaling distortion and reduce the amount of calculation.In order to further accurately extract the content and information of scaled and distorted images,and obtain better classification results and recognition accuracy,this paper also studies the bag of words model(BOF)and its improved algorithm.Several aspects such as point extraction,feature descriptors and selection of SVM classifier kernel parameters have been studied in depth.Firstly,this paper studies the method of feature descriptor extraction,and gives the principles and steps of two different extraction methods which SIFT algorithm and SURF algorithm.Based on this,a match between the visual dictionary and the spatial pyramid is constructed to improve the classification and recognition effect of the scaled image and improve the recognition accuracy.Secondly,this paper compares the kernel parameters of four SVM classifiers,and selects the pyramid histogram intersection with the highest classification accuracy kernel as the kernel parameter of the classifier.Finally,this paper combines the improved bicubic interpolation method with the bag-of-words model for classification of scaled images.According to the test results,the improved image classification accuracy of the proposed algorithm combined with the improved bicubic interpolation method and the bag-of-words model is higher than several other classification methods.