Research On Garment Style Recognition Method Based On Digital Image Processing

Using digital image processing technology to identify garment styles from the garment image,it has a wide application prospect in garment consumption analysis and assistant fashion design and identification.Garment style features are mainly reflected by the contour characteristics of garment.In the field of garment style recognition,the research on garment contours feature extraction and classification,the main methods include: the extreme learning machine classification based on wavelet Fourier descriptors,and the Euclidean distance classification based on fusion features.However,these methods also have some shortcomings,such as wavelet Fourier descriptor similarity discrimination is more complex,ELM classification of poor adaptability and Euclidean distance discrimination efficiency is low.At present,there is no effective method for garment contours feature extraction and classification.To address the problem above,this paper explores a new and effective method for the garment image recognition based on digital image processing technology.For the main feature of contour feature extraction and its classification method,two kinds of different technical schemes are proposed: template matching method based on contour curvature feature points and SVM classification method based on contour Fourier descriptor.After the specific design and implementation,the optimal method is obtained by comparing the results of style recognition..The main work of this paper is summarized as follows:Firstly,this paper analyzes the background and present situation of the research,and puts forward the topic by analyzing the shortcomings of the current research work.Secondly,because there is no internationally recognized image database,this paper creates a new sample database.This paper designs a preprocessing scheme based on gray-scale enhancement for the condition that the traditional garment image preprocessing algorithm based on edge detection had poor effect on garment contour extraction.By appropriately stretching the gray scale interval,the contrast between the garment image and the background is enhanced.After binarization and a series of morphological processing,the garment contour are obvious.The contours detected by the canny operator are Fourier filtered in the frequency domain.Finally we get the contour of the garment.The results show that the contour of the garment is smooth and can reduce the texture noise effectively without losing the original structure.Thirdly,according to the current contour feature extraction technology is more complex,and its classification method of low efficiency and poor adaptability,this paper presents two schemes.Scheme 1: Hausdorff template matching method based on contour curvature feature point feature.Scheme 2: SVM classification method based on contour Fourier descriptor feature.In the scheme 1,the characteristic of contour curvature points can describe the curve characteristics of garment contour,which is simple and intuitive;and the improved mean Hausdorff distance is better.In the scheme 2,Fourier descriptor expression profile of global shape feature,and his each component has a certain physical meaning,and similarity discrimination is simple;SVM classification and generalization ability is outstanding.We achieved the above program and verified the feasibility,through detailed design and Matlab programming.Finally,through the comparison of the experimental results of scheme 1 and 2,it was found that the recognition accuracy of the scheme 2 is about 95.9% for a image database of 650 garment images,about 17% higher than the scheme 1;The scheme 2 to identify each image takes about 3.6 milliseconds,about 20 times faster than the scheme 1;However,the scheme 1 has a better effect on the identification of garment style with prominent contour feature,and the feature description is more intuitive.In terms of the recognition effect on the individual and the whole,the scheme 2 is identified as a superior scheme because of its advantages.And the advantage of scheme 2 is further confirmed by comparing scheme 2 with other contour feature extraction methods(Hu invariant moment,fusion feature)and other classification methods(extreme learning machine,BP neural network).This paper presents and implements the computer image recognition scheme of garment styles,which can automatically recognize garment styles in garment images.The preprocessing scheme can effectively eliminate the interference of texture noise.The contour feature extraction and its classification scheme are characterized by strong feature description,simple calculation,accurate pattern recognition,real-time and adaptability.The proposed method has certain reference value for the practical application of the garment style recognition,laying a certain foundation for the development of the clothing style automatic identification system for complex garment images.