Embossed surfaces are widely used in many electronic products to enhance the appearance and user experience. However, there are inevitably some defects in the production process. They are difficult to detect by traditional detection algorithms because of the irregular textures on the surface, which are easily affected by illumination conditions. To achieve robust defect detection for logotype printed on the embossed surface, we proposed a novel defect detection method called multiple paired pixel consistency (MPPC). Firstly, we propose a consistency measure for high consistency pixel pairs to realize a robust defect-free model. Secondly, based on this model, we design a measure for judging defective pixel or defect-free pixel. Experimental results with some real-world defective images demonstrate that our approach can achieve state-of-the-art accuracy in real industrial applications.

Automatic logo detection is a key function for many applications. Most existing logo detection methods are based on strong object level bounding box annotations, annotated either manually or automatically in synthesized images. However, in real applications, there are thousands of different types of logos with clutter background and varying sizes in images, that makes supervised detection methods less applicable because it is a labor-intensive task to do bounding box annotations. On the other hand, some weakly supervised learning-based methods have been proposed, their experimental results are not promising. In this paper, we propose weakly supervised methods for logo detection based on a dual attention dilated residual network (DRN) containing spatial and channel attention mechanisms using image-level labels instead of bounding box annotation for training data. The incorporated spatial attention module computes attention weights which are useful in predicting the spatial location in an image. Channel wise feature maps are used to emphasize the dependency of the channel in the global feature map and help to classify the logos into different categories. The use of attention-based mechanisms with DRN improves classification accuracy (around 4%) and considerably increases localization accuracy by more than 4% over a regular dilated residual network architecture.

The expression of flames using computer graphics is frequently required as a familiar phenomenon. Various colors, including red and blue, can be seen in real flames. However, the combustion reaction in a real flame is too complex, involving thousands of chemical reactions to render interactive simulations. In this study, the authors constructed a novel combustion reaction model for expressing realistic flames with simplified chemical reactions of complex combustion. To achieve interactive physicochemical simulations of flames, fluid physics simulations were applied using the position-based fluids method. In the proposed combustion model, the combustion reaction is represented as five dominant chemical reactions. Opposed to the conventional methods, the proposed model can reproduce the blue and red regions of the flame. The proposed flame simulation method can be operated at approximately 20 frames per second or more, making interactive rendering possible.