1．烟雾纹理和边缘多尺度特征 - 中国图象图形学报.doc

中图法分类号：TP391 文献标识码：A 文章编号：1006-8961(200 ) - - 论文引用格式： 采用金字塔纹理和边缘特征的图像烟雾检测方法 李红娣1，袁非牛2 1江西财经大学图书馆，南昌3300132.江西财经大学信息管理学院，南昌330032 1 关键词：边缘方向直方图；局部二元模式；支持向量机；图像烟雾检测 Image Based Smoke Detection Method Using Pyramid Texture and Edge Features Hongdi Li1, Feiniu Yuan2 1. Library, Jiangxi University of Finance Economics, Nanchang 330013, Jiangxi, China; 2. School of Information Technology, Jiangxi University of Finance Economics, Nanchang 330032, Jiangxi, China Abstract: Objective: Image based smoke detection methods have many advantages over traditional point based smoke sensors, such as fast response, non-contact. But it’s still a challenging task for existing methods to accurately detect smoke from images due to large variance of smoke shape, color and texture. Method: To improve recognition accuracy, we extract features of pyramidal textures and edges to propose a novel image based smoke detection method. In this method, an image is first decomposed into an image pyramid, and then we extract Local Binary Patterns (LBP) and Edge Orientation Histogram (EOH) from each layer of the image pyramid, which are called Pyramidal LBP (PLBP) and Pyramidal EOH (PEOH), respectively. We adopt different pooling schemes to generate sequential histograms of PLBP and PEOH that represent smoke textures and edges, respectively. At last, we concatenate these histograms together to form smoke feature vectors and use support vector machine (SVM) for training and classifying. Since image pyramids contain scale information, our pyramidal texture and edge features have certain scale-invariance. Results: Experimental results show that the method obtains detection rates above 94% and false alarm rates less than 3.0% on our large image datasets. Conclusion: Texture and edge features extracted by our method have certain illumination and scale invariance, and experiments show these features have good discriminations and generalization performance for smoke detection. Keywords: Edge Ori