Chinese Conference on Pattern Recognition 

About: Chinese Conference on Pattern Recognition is an academic conference. The conference publishes majorly in the area(s): Feature extraction & Feature (computer vision). Over the lifetime, 1522 publications have been published by the conference receiving 4566 citations.

Deep Convolutional Neural Network with Mixup for Environmental Sound Classification

Abstract: Environmental sound classification (ESC) is an important and challenging problem. In contrast to speech, sound events have noise-like nature and may be produced by a wide variety of sources. In this paper, we propose to use a novel deep convolutional neural network for ESC tasks. Our network architecture uses stacked convolutional and pooling layers to extract high-level feature representations from spectrogram-like features. Furthermore, we apply mixup to ESC tasks and explore its impacts on classification performance and feature distribution. Experiments were conducted on UrbanSound8K, ESC-50 and ESC-10 datasets. Our experimental results demonstrated that our ESC system has achieved the state-of-the-art performance (83.7\(\%\)) on UrbanSound8K and competitive performance on ESC-50 and ESC-10.

Multi-stream Deep Networks for Person to Person Violence Detection in Videos

Abstract: Violence detection in videos has numerous applications, ranging from parental control and children protection to multimedia filtering and retrieval. A number of approaches have been proposed to detect vital clues for violent actions, among which most methods prefer employing trajectory based action recognition techniques. However, these methods can only model general characteristics of human actions, thus cannot well capture specific high order information of violent actions. Therefore, they are not suitable for detecting violence, which is typically intense and correlated with specific scenes. In this paper, we propose a novel framework, i.e., multi-stream deep convolutional neural networks, for person to person violence detection in videos. In addition to conventional spatial and temporal streams, we develop an acceleration stream to capture the important intense information usually involved in violent actions. Moreover, a simple and effective score-level fusion strategy is proposed to integrate multi-stream information. We demonstrate the effectiveness of our method on the typical violence dataset and extensive experimental results show its superiority over state-of-the-art methods.

A Fast 2D Otsu Thresholding Algorithm Based on Improved Histogram

Abstract: Otsu adaptive thresholding is widely used in classic image segmentation. Two-dimensional Otsu thresholding algorithm is regarded as an effective improvement of the original Otsu method. To reduce the high computational complexity of 2D Otsu method, a fast algorithm is proposed based on improved histogram. Two-dimensional histogram is projected onto the diagonal, which forms 1D histgram with obvious peak and valley distribution. Then two-dimensional Otsu method is applied on a line that is vertical to the diagonal to find the optimal threshold. Furthermore, three look-up tables are utlilized to improve the computational speed by eliminating the redundant computation in original two-dimensional Otsu method. Theoretical analysis and experimental simulation show that the proposed approach greatly enhances the speed of thresholding and has better immunity to Salt and Pepper Noise.

Moving Cast Shadow Detection of Vehicle Using Combined Color Models

Abstract: Moving cast shadow detection of vehicle is important to vehicle detection and tracking. In this paper, a novel moving cast shadow detection approach using combined color models is proposed. Firstly, the ratio of hue over intensity in HSI color model is used to detect the bright object pixels in foreground regions. Secondly, we employ the theory of photometric color invariants in c1c2c3 color model to distinguish the dark (similar to shadow) and colorful object pixels from shadow pixels. Finally, to improve the accuracy of shadow detection, post processing is used to correct failed shadow and object detection. Experimental results indicate that the proposed method can detect moving cast shadows accurately.

Chinese Handwriting Recognition Contest 2010

Abstract: Chinese handwriting recognition remains a challenge. Research works have reported very high accuracies on neatly handwritten characters yet the performance on unconstrained handwriting remains quite low. To promote the recognition technology, new databases of unconstrained handwriting have been constructed for academic research and benchmarking. This paper reports the contest results of online and offline handwritten Chinese character recognition using the new generation of databases, targeting 3,755 Chinese characters of the GB2312-80 first level set. Nine systems from four groups were submitted for evaluation. The best results are 92.39% accuracy for online character recognition and 89.99% accuracy for offline character recognition. Detailed analysis of results on data of different writers reveals the diversity of writing quality. The future contests will consider continuous script recognition as well as isolated character recognition.