朱晨捷 鄭靈鳳 葉煒 朱蓉



摘 要: 提出了一種針對火災圖像進行處理的特征提取及識別方法。首先,通過建立顏色直方圖提取火災圖像與非火災圖像的顏色特征;然后,利用尺度不變特征變換算法計算兩類圖像的局部特征,并利用主成分分析法對兩類圖像特征進行降維處理,再針對降維處理后的圖像特征采用K均值聚類算法進行計算;最后,針對測試圖像庫中的圖像數據,經過顏色直方圖初判、局部特征與聚類中心對比等步驟獲得識別結果。該方法能夠將火災圖像有效、快速地識別出來,以達到及時報警的效果。
關鍵詞: 火災圖像識別; 顏色直方圖; 特征提取; 尺度不變特征變換; 主成分分析法; K均值聚類
中圖分類號:TP39 文獻標志碼:A 文章編號:1006-8228(2015)12-26-04
Research on feature extraction and recognition method for fire image
Zhu Chenjie, Zheng Lingfeng, Ye Wei, Zhu Rong
(College of Mathematics and Information Engineering, Jiaxing University, Jiaxing, Zhejiang 314001, China)
Abstract: In this paper, a method of feature extraction and recognition for fire image is presented. The color features of fire images and non-fire images are extracted respectively via the establishment of color histograms. Then, the local features of these two kinds of images are computed by using SIFT (Scale Invariant Feature Transform) algorithm, and reduce the dimension by using PCA (Principal Component Analysis) method and those processed local features are calculated by K-means clustering algorithm. The recognition results are obtained by the prejudgment of color histograms and the comparison of local features with clustering centers. It is proved by experiments that the proposed method can recognize the fire image effectively, and can achieve a good result in early-warning.
Key words: fire image recognition; color histogram; feature extraction; Scale Invariant Feature Transform (SIFT); Principal Component Analysis (PCA); K-means clustering
0 引言
由于每年發生的不同程度的火災給人們帶來了巨大的經濟損失,因而對火災監控和火災發生后及時報警進行研究一直是研究者關注的重點。傳統的火災報警是通過煙霧、溫度來進行檢測,相比基于圖像的檢測方法準確率低。因此,本文提出了一種基于圖像內容的火災識別方法。
通常,火災圖像上會呈現一些顯著的特點,從圖1給出的這些火災圖像中可見,火焰區域多數呈橘紅色,煙霧區域多數呈灰白色;從紋理上看,火災區域與其他區域的區別比較明顯。
本文在深入分析火災圖像顏色和紋理特性的基礎上,提出了一種基于顏色直方圖、尺度不變性特征變換(Scale Invariant Feature Transform,簡稱為SIFT)算法和K均值聚類算法的火災圖像特征提取與識別方法。……