Fabric defect detection is a crucial problem in the production of textile industry. Nowadays, large number fabric algorithms based on deep learning have been proved to excellent results, but performance often depends large-scale labeled datasets. However, it very laborious and costly label pixel-level samples industrial production. Although semisupervised method can reduce labeling workload, still challenging task due diversity defects low contrast between background. To address these...

In this paper, we propose a novel method for the detection of microcalcifications using mathematical morphology and support vector machine (SVM). First, contrast in original mammogram was improved by gamma correction two carefully desig

Paper currency recognition is widely applied in many fields such as bank system and automatic selling-goods system. How to extract highly qualified monetary characteristic vectors from image an important problem needing be solved now. It a key process select original information with noises uneven gray. This article, aiming at the specialties of RENMINGBI(RMB) image, puts forward method using linear transform gray diminish influence background order give prominence edge image. Then obtained...

Recovering high-quality inscription images from unknown and complex noisy is a challenging research issue. Different natural images, character pay more attention to stroke information. However, existing models mainly consider pixel-level information while ignoring structural of the character, such as its edge glyph, resulting in reconstructed with mottled local structure damage. To solve these problems, we propose novel generative adversarial network (GAN) framework based on an edge-guided...

In this paper we propose a gait recognition method with dynamic energy image (DGEI) and manifold learning. First present new feature-dynamic which can reflect the variance parts of motion body better characterize features. Secondly in order to preserve principal discriminant components, use PCA LPP discover low-dimensional high feature space, characteristics DGEI are well preserved. Lastly simple vote rule Dempster-Shafer (D-S) evidential theory used as fusion strategy for fusing multi-views...

Inverse halftoning acting as a special image restoration problem is an ill-posed problem. Although it has been studied in the last several decades, existing solutions can’t restore fine details and texture accurately from halftone images. Recently, attention mechanism shown its powerful effects many fields, such processing, pattern recognition computer vision. However, not yet used inverse halftoning. To better solve of detail halftoning, this paper proposes simple effective deep learning...

Retinal image vessels can reflect different states of many diseases, which are often used for the registration and mosaic retinal images. Therefore, it is very important to accurately extract blood vessels. The paper firstly analyzes characteristic images, secondly gray mathematical morphology theories smooth strengthen images in order remove background enhance brightness vessels, finally extracted with entropy threshold method. Compared result segmented by manual experts, our approach get...

In video human action recognition of the continual motion is a difficult point for application. The method based on latent-dynamic condition random fields presented. By star form distance descriptor body contour, pose extracted. Then in continuous sequences building model LDCRF shows mapping relation between feature and semantics. Comparing with traditional CRF HCRF, by designing affiliation latent pose, implements modeling internal external movement feature. experiment, Weizmann database...

Human action recognition is an important issue in the field of computer vision. But existing models put more emphasis on single scale and little attention multi-scale multi-action mode motion. With aim at this problem, paper presents a human motion method using condition random model. At first, trajectory movement, body posture characteristics as well limb movement are considered order to extract feature. Then Conditional Random Fields model proposed for recognition. The can take full...

The denoising of a natural image corrupted by noise is classical problem in signal processing. A new method for discussed. bivariate shrinkage function based on the dependency between wavelet coefficients derived from Bayesian maximum posterior (MAP) estimation theory and applied to denoising. performance compared with that conventional soft thresholding technique. Experimental results show satisfying suppression, preserving edges details.

Human activity analysis based on video is a hot research topic in the field of computer vision, which segmentation human sequence fundamental question. In this paper, we present new unsupervised method for sequence. The main idea that most dramatic point, regarded as split detected by gaussian model according to action's mutation video. Experimental results show feasibility and effectiveness algorithm paper.

This paper presents a simple yet effective 3D posture carving reconstruction method for human action recognition. The idea is inspired by the observation that could carve object according to few number of 2D images collected from multiple cameras. First, sequences are captured simultaneously 3 web Then normalized binary silhouettes obtained series preprocessing steps. Finally, in virtual coordinate reconstructed different views. Experimental results show effectiveness proposed approach.