Research On The Key Technology For Furnace Flame Feature Extraction

Recently, with the development and progress of multimedia technology changingrapidly, people are exposed to a large number of image information in daily life,which promote the research and development of image processing technology. Withdifferent contents of rich information contained by various images themselves, imageprocessing technology has been developing rapidly in all fields of society, and thefurnace flame image processing technique is an important application domain. Videocapture of the flame image by CCD camera, and through the image acquisition cardtransform into the digital image which can be processed by computer, and transformqualitative analysis of furnace flames into the quantitative analysis of flame images.At the same time, by processing of the furnace flame image, make the computermonitoring furnace flame combustion conditions and measurement of the furnacecombustion temperature possible. Therefore, the furnace flame image processing technology is concerned about by more and more researchers at home and abroad, andbecome a research hotspot in the field.Analysis of the characteristics of the image is one of the key factors in imageanalysis, through the description and expression of image feature, the extracted imagecontains the original property or attribute, so as to lay a foundation for imageanalysis or identification. The main research contents of this paper is that applicationof image processing technology extracts the characteristic information of the furnaceflame image, including color feature, shape feature, texture feature of the flameimage. According to the characteristics of the furnace flame images, put forward afurnace flame image feature extraction method based on hierarchical salient points.First of all, Haar wavelet transform is used to calculate salient value, under atop-down approach from quadtree data structure, the algorithm keeps the most salientpoints in each quadrant according to the percentage of saliency values in the ancestornode. On the basis of this, in order to make the extracted points showing moresignificant characteristics of the edge, a hierarchical adaptive method to extract salient points was proposed. First the block difference of inverse probabilities model wasused to change the original image into block difference of inverse probabilities image.On the basis of this, made the BDIP image into Haar wavelet transform, calculated thesalient value of two-dimensional image. According to the furnace flame image underdifferent condition with the characteristics of complexity and difficult classification.This paper put forward the method of Hierarchical Adaptive Fast K-means(HAFKM)classifying and clustering for image database. According to the proposed hierarchicalstrategy， HAFKM builds an unbalanced clustering tree, and through the method ofadaptive CEC (Cluster Evaluation Criterion) determines every subtree branchnumb-er except the root node, and in each layer of the clustering tree uses a proposedcost-function clusters on the color level histogram directly, and then database can fastcluster in the whole tree.In the process of algorithm analysis, By compared this algorithm withtraditional classical algorithm under the same experimental conditions, proved that inthe process of extraction salient points using this algorithm, this method overcomesthe shortcoming of traditional wavelet transform to extract too many significant pointsand local points gather easily, at the same time the method is robust and the extractedsalient points provide efficient retrieval performance. In the algorithm of HAFKM,HAFKM through the layered cluster in the unbalanced tree, and through CECdetermining the correctnumber of clusters, can realize the classification of databasefast and efficiently in the end.