Skirt Image Recognition Based On Feature Extraction And Convolution Neural Network

By processing some key points or features of the input garment images,the computer can extract the important features of garment images for garment recognition,which is of great significance in the field of garment e-commerce.However,the current methods are not ideal in dealing with the details of clothing.There are some problems,such as inaccurate selection of structural feature points of internal details and ignorance and confusion of process details.For example,details such as hemispheric pockets and cuffs are not recognized.Therefore,this paper explores a method of feature extraction combined with convolution neural network to recognize hemispheric skirt image.According to the structure characteristics of hemispheric skirt,the realization and verification of feature extraction and recognition scheme for hemispheric skirt are carried out.This topic mainly includes the following contents and conclusions:1.Establishing a data set including four categories of bust skirts,which included external contours,internal details,style and pattern.The crawler technology was used to collect 21000 bust skirt images,which laid a foundation for subsequent image recognition and retrieval.2.Using edge extraction to extract the features of the outer contour of the halflength skirt image,the effects of Canny operator and Sobel operator are compared and the best parameters are obtained.Among them,the parameter value of Canny operator is 80,and that of Sobel operator is 1,which was the best parameter value of the algorithm.3.Corner extraction was used to extract the interior details of half-length skirt image,and the effect of feature extraction by Harris operator and Fast operator is compared,and the best parameter value is obtained.Among them,the parameter value of Harris operator is 1,and that of Fast operator is 30,which was the best parameter value of the algorithm.4.The convolution neural network was used to train a large number of pictures of the hemispherical skirt pictures which have been extracted from the features.The speed of the training was increased by about 37%,and the average speed of the recognition of a single picture was 0.0714 seconds.5.Using TensorFlow framework of convolutional neural network to recognize hemispheric skirt image,the average recognition rate is 93.87%,which was about 10%higher than that of support vector machine.In conclusion,this topic proposes and realizes the feature extraction and recognition of hemispheric skirt style.Contour extraction can extract the edge information of hemispheric skirt completely,corner extraction can accurately locate the details of hemispheric skirt.The proposed method can provide design inspiration for fashion designers,and provide reference and basis for clothing e-commerce platform.