In order to solve the insufficiency of training data when deep learning technology is applied to surface defect detection task, a surface defect generation algorithm based on generative adversarial network (GAN) was proposed to enhance training sample data. First, a U-shaped convolutional network was designed, and a spatial adaptive normalized structure was introduced to control the mask image to generate the defect shape, and the network from defect-free image to defect image was completed. Second, a multi-layer convolutional discriminant network is designed to extract adversarial feature of the real samples and generated samples. Finally, the adversarial training loss was designed and the generative network adversarial training was completed. Through quantitative contrast experiment, it is proved that the segmentation network has better segmentation results than without data augmentation after using the surface defect generation algorithm to generate data for data augmentation.
Automatic license plate detection and recognition (ALPDR) in natural scene is a useful but difficult task as the all-weather and variety of lighting conditions. Though deep learning based ALPDR methods can achieve much higher recognition rate, it needs a large number of human-labelled samples to train the deep neuron network. In this paper, we propose a method to generate synthetic data based CNN ALPDR to avoid manually labelling lots of data and stabilize training. First, our data engine generates 100K synthetic car license plates to simulate real scene and train networks. Then, we design a recognition network to predict all characters holistically, avoiding the character segmentation. Some real scene data sets are employed to validate the effectiveness of our presented method. The accuracy of our ALPDR system is 91.18% and 95% in toll station dataset and 94.2% in traffic surveillance dataset.
Optical character recognition (OCR) in complex scenes, particularly in industry environment, is a challenging problem that has received a significant amount of attention. A unified model for different types of character in different production lines is needed. In this paper, we propose a unified framework to classify characters using convolutional neural network (CNN) to satisfy the two main requirements in industrial OCR, the high recognition rate and less training time by combining the representational power of multi-layer neural networks together with multi-stage features. In the model, there are three CNNs, two with multi-stage features and one with deeper layers, which can be used to extract different fonts and types characters in different complex background. The results in experiments demonstrate the efficiency with high recognition rate and less training time in complex industrial environment.
To obtain information on the outer surface of a cylinder object, we propose a catadioptric panoramic imaging system based on the principle of uniform spatial resolution for vertical scenes. First, the influence of the projection-equation coefficients on the spatial resolution and astigmatism of the panoramic system are discussed, respectively. Through parameter optimization, we obtain the appropriate coefficients for the projection equation, and so the imaging quality of the entire imaging system can reach an optimum value. Finally, the system projection equation is calibrated, and an undistorted rectangular panoramic image is obtained using the cylindrical-surface projection expansion method. The proposed 360-deg panoramic-imaging device overcomes the shortcomings of existing surface panoramic-imaging methods, and it has the advantages of low cost, simple structure, high imaging quality, and small distortion, etc. The experimental results show the effectiveness of the proposed method.
In this paper, we propose a novel cascade detection algorithm which focuses on point and line defects on TFT-LCD. At the first step of the algorithm, we use the gray level difference of su-bimage to segment the abnormal area. The second step is based on phase only transform (POT) which corresponds to the Discrete Fourier Transform (DFT), normalized by the magnitude. It can remove regularities like texture and noise. After that, we improve the method of setting regions of interest (ROI) with the method of edge segmentation and polar transformation. The algorithm has outstanding performance in both computation speed and accuracy. It can solve most of the defect detections including dark point, light point, dark line, etc.
Automatic inspection takes a great role in guaranteeing the product quality. But one of the limitations of current inspection algorithms is either product specific or problem specific. In this paper, we propose a defect detection method based on three image features fusion for variety of industrial products surface detection. The proposed method learns sub-image gray level difference, color histogram and pixel regularity of qualified images off-line and test the images based on the detection results of these three image features. It avoids the feature training of defect products as it is difficult to collect large amount of defect samples. The experimental results show that the detection accuracy is between 93% and 98% and the approach is efficient for the real time applications of industrial product inspect.
Aiming at automatic visual inspection of texture surface, a texture surface defect detection method is proposed based on statistical feature of subimage. The proposed method only uses a simple image feature, gray level difference of subimage without image enhancing to detect defects on texture surface directly, avoid the feature computation of high dimension space and the learning process of large numbers of defective and defect-free similar images, which is nonsupervised detection and improving algorithm efficiency. A variety of texture surfaces from industrial manufacture materials are chosen to conduct experiments. Detection time is about few seconds and accuracy is 93.6%. Experiment results prove the proposed method can online detect various texture surface defects effectively.
This paper presents a framework for robustly and accurately computing the visual hull of a real object from images
sequences. Unlike most existing volumetric based approaches, level set deformable model is utilized in our system to
drive the surface from a sphere smoothly recovery the shape of the real object. The algorithm represents the object's
surface implicitly as the zero level set in uniform grid and the visual hull computation problem is translated into a forces
computation problem. The deforming surface evolves under the internal and external forces according to the silhouettes
and smoothness constrains. Snake deformable model is applied as a refinement step to improve the quality of mesh and
reduce the total computing time. This classical and geometric mixed deformation model can easily and naturally changes
the topology of the surface and need not add any extra measurement to avoid mesh confusion. The experiment results
turns out that the final mesh have higher precise and smoothness than the traditional volumetric based approaches.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.