A Simple Hierarchical Clustering Method for Improving Flame Pixel Classification

Improving the Quality of Color Image Segmentation Using Genetic Algorithm

“…Our proposal uses the algorithm HCM presented in [13] (see Algorithm 1) for simplifying color images in order to decrease the number of colors. This method needs six parameters [13]: (i) original image; (ii) number of colors, which represents the number of representative colors that will be identified; (iii) outlier threshold, which represents the smaller permitted connected component size in terms of color frequency; (iv) a color space, which is the basis for graph creation values; (v) a distance measure, which is used to define the weight of graph edges; and (vi) number of nearest points considered during the graph creation. The computation time is directly related to the number of colors and, of course, to the adjacency relation of the graph, which will influence the graph size.…”

“…Color image segmentation aims to group image regions using some criterium, however the choice of a strategy for grouping is a difficult task, and it is dependent on the application domain. One approach to cope with this problem is to model an image as a grid graph whose vertices correspond to the pixels and the edges connect the nearest neighbor pixels and their labels represent a dissimilarity measure computed from the connected pixels [4,13] followed by the computation of a minimum spanning tree (MST). The first appearance of this tree in pattern recognition dates back to the seminal work of Zahn [15].…”

“…In [7] two methods were proposed for color image segmentation, these methods also are MST-based and are called Hierarchical Euclidean Minimum Spanning Tree (HEMST) and Maximum Standard Deviation Reduction Clustering Algorithm (MSDR). In fact, image segmentation methods may be considered as a clustering problem that are usually handled by MST and K-means approaches (see [3,5,10,12,13,14] for more informations).…”

“…In [13], a simple method was proposed for simplifying color image using a minimum spanning tree approach for improving fire pixel classification. Even if the results for this application were quite good, the tuning of parameters is a real challenge.…”

“…In this paper, we present a method for improving the quality of image segmentation using a genetic algorithm. Our proposal is based on the Hierarchical Clustering Method (HCM) presented in [13] and, therefore, we named it Genetic Improvement of Hierarchical Clustering Method (GHCM). The main contribution of this work is a graph-based approach for image segmentation, which uses a genetic algorithm to optimize the quality assessment according to some goodness method [17].…”