Multi-view Characterization And Assessment Of Fabric Appearance Smoothness Based On Sparse Coding

As the comprehensive effect of physics and chemistry, the fabric surface produced durable creases during home or industrial launder ing and caring, destructing the aesthetic appearance and performance of textile and clothing. Fabric appearance smoothness reflects shape retention of material, widely used to assess the performance of materials, fabric manufacture and finishing, effect of home/industry laundering and caring, etc.. Artificial subjective method and existed objective methods were difficult to meet the demand of objective, accurate detection and subjective-objective consistent principle. In this paper, the sparse coding algorithm and multi-view reconstruction methods were proposed to fulfill the objective assessment of fabric smoothness, which combine with the image analys is technology and human visual perception based on the basic theory of textile and garment. The proposed method has distinctive features as new quality evaluation technology and enhances the traditional industry.This paper simulated the mechanism of human visual perception and information processing in artif icial assessment. Three aspects including local feature extraction and matching of fabric images, reconstruction in multi-views, sparse coding and auto rating based on depth image were studied. The main research contents and conclusions were summarized as follows:(1) Analysis of the advantages and disadvantages of artificial visual assessment. As to reveal the advantages and defects of subjective assessment, define the application scope of subjective inspection, and clarify the research direction of objective rating, the human visual perception and information processing mechanism, assess condition, method and standard replicas were discussed in this paper. Besides, the subjective visual rating bias, the differences between individuals and their visual impact of factors were systematically analyzed based on the eye tracking experiment and image processing technology.Research on the theory of biological vision perception and subjective assessment of artificial vision shows that the human eye can perceive the three-dimensional shape of the object and has a fast, low-cost, multi-feature expression of natural images. Artif icial vision rating was arranged in a specific experimental environment by the trained inspectors and present a comprehensive evaluation compared with the standard replicas. The experimental results of eye tracking showed that the artificial assessment had less deviation with the pure white fabric and the result was relatively stable. The subjective rating inspectors were adhere their own cognitive criteria as own differences were small, but individual differences was signif icant. The impact of color on the assessment of fabric with color or texture shows quite significant which was caused visual attention differences. The more the visual coverage area deviates from the wrinkle position, the bigger the assess deviation showed. The more the visual focus is dispersed, the greater the inter-individual rating dispersion. Results showed that artificial grading is only partially applicable to pure white fabric ratings and not suit to assess fabrics with color or texture. It is concluded that the subjective rating is susceptible to the interference of the color and texture of the fabric surface, resulting in the distortion of the grading result and the individual deviation.(2) Development of multi-view image acquisition device. Based on artificial assess habits and convenience of inspection, the straightedge, rotable and portable multi-view image acquisition device were developed after a series of comparison of platform, environment light source, imaging device, etc.. The multi-view imaging device can fulfill automatic rotation, binocular image synchronous acquisition, transmission, and storage.The device consists of image acquisition platform, light source unit, imaging, real-time image transmission and other units, to achieve automatic camera rotation, image synchronization acquisition, transmission and storage functions. The weighted average method was used to gray the image, and Harris corner detection was used to determine the effective area of the creased fabric, and the adaptive mean filter was used to eliminate the noise. Besides, the acquired images were preprocessed by effective region segmentation, denois ing and greying, which offered the effective image information for calibration, matching, stereo reconstruction and depth mapping. On the basis of the results, The image preprocessing provides the reliable bas ic information for the later image correction and matching, stereo reconstruction and depth mapping.(3) Extraction and matching of wrinkle fabric. In order to avoid the interference of texture and color, this paper proposed the binocular stereo reconstruction method using image local feature descriptors and similar ity to fulfill double images matching. Firstly, inner and outer parameters of two parallelly placed cameras were calibrated by checkerboard calibration algor ithm, and images’ distortion was corrected and the rows were strictly aligned to narrow matching range. Secondly, Scale Invariant Feature Transform(SIFT) combined the region growing mechanism and k-d tree storage strategy, to fulfill the image matching according to Normalized Cross Correlation(NCC) algor ithm. Finally, height of wrinkle image was calculated by coordinate system transformation.In this paper, 6 types of fabrics with the different appearance texture are selected to verify the proposed algorithm. The experimental results show ed that the wrinkle shape of the fabric can be reconstructed partially by the binocular stereo vision algorithm, which avoids the color and texture interference. Binocular stereoscopic vision can effectively avoid the interference of color and texture. However, due to the influence of perspective and matching precision, some regions have perspective distortion which leads to the omission of matching in some areas.Binocular stereo vision could effectively avoid the interference of color and texture during reconstructing, but appearance local missing matching as perspective distortion.(4) Multi-view reconstruction and depth mapping. In this paper, we proposed a new method to solve the problem of missing image matching, using rotable binocular cameras to capture images from different angles. Different angles of images can be transferred to initial pos ition by checkerboard calibration. Secondly, the iterate method ‘point to module’ was utilized to fulfill cloud points matching. Height information then was mapped as a grey image which called ‘depth image’ as the basic information of classification.In the experiment part, the camera angle transformation, point cloud denoising, point cloud registration, optimization and visualization algor ithm were developed to realize multi- purpose 3D reconstruction. The results show ed that the multi-view fabric reconstruction algor ithm can multi-view reconstruction can effectively fusion point clouds and reconstruct the space morphology of wrinkle fabric. The proposed method can effectively integrate point cloud at different viewpoints in the same coordinate system by multi-view image information, effectively rebuild a complete space morphology of wrinkled fabric.(5) Image coding and objective assessment of fabric appearance smoothness. In the part of image coding and rating, features of depth image were detected and coding by sparse coding algor ithm. Sparse coding algor ithm referenced the information processing mechanism of mammalian reception of vision cells, which simulates the process of human inf ormation processing and judgment. The sparse coding algorithm is used to train and code the underlying features of the fabric depth image, and the visual appearance dictionary of fabric visual f latness is constructed to characterize the fabric folding profile. Smoothness classification was made by multiple classification strategy of support vector machine.In the experiment, 300 types of fabrics with different surface appearance were selected and the depth images of the fabrics were designed using 3D reconstruction technique proposed in this paper. The experimental results show that the sparse coding algorithm can reflect the essential characteristics of the image, and the subjective and objective assess consistency showed more signif icant improvement than the Bo W and SPM algorithms. When the number of training samples is 90, the correct classification rate reach 95.3%, singer depth image rating time is less than 3 seconds, accuracy class reach 0.1, which can assess the smoothness grade in a short time.Innovations include:(1) it’s the first time to refer artificial visual perception to develop a multi-view and portable image acquisition device, which utilized the eye disparity.(2) Innovatively integrated SIFT local invariant features, and combined with regional growth mechanism and registration method of ‘point-to-surface’ to reconstruction the morphology of wrinkle fabric, which can valuable avoid the inf luence of color or texture on fabric for inspection.(3)Simulating fast, low-cost expression mechanism of biological visual neural for natural images, local features were extracted as visual ‘points of interest’ of the under characteristics, utilizing the sparse coding algor ithm to construct the ‘visual dictionary’ alternative mathematical indexes for characterizing wrinkle fabric. Research of the assessment of fabric appearance smoothness improved the accuracy and stability of fabric appearance performance. Research of the assessment of fabric appearance smoothness improved the accuracy and stability of fabric appearance performance, and has an important practical value for the assessment of textile quality inspection, laundry/care equipment and product performance.