## An Effective Color Quantization Method Based on the Competitive Learning Paradigm

### BibTeX

@MISC{Celebi_aneffective,

author = {M. Emre Celebi},

title = {An Effective Color Quantization Method Based on the Competitive Learning Paradigm},

year = {}

}

### OpenURL

### Abstract

Abstract — Color quantization is an important operation with many applications in graphics and image processing. Most quantization methods are essentially based on data clustering algorithms one which is the popular k-means algorithm. A common drawback of many conventional clustering algorithms is the generation of empty clusters (dead units). In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm [1] to the problem of color quantization. In contrast to the conventional batch k-means algorithm, this competitive learning algorithm requires no cluster center initialization. In addition, it effectively avoids the dead unit problem by utilizing a simple cluster splitting rule. Experiments on commonly used test images demonstrate that the presented method outperforms various stateof-the-art methods in terms of quantization effectiveness.

### Citations

1266 |
An algorithm for vector quantizer design
- Linde, Buzo, et al.
- 1980
(Show Context)
Citation Context ...hiyama and Arbib’s competitive learning algorithm, respectively. Section 4 presents the experimental results, whereas Section 5 gives the conclusions. 2. Batch K-Means Algorithm The k-means algorithm =-=[30]-=- is inarguably one of the most widely used methods for data clustering [31], [32]. Given a data set X = {x1, . . . ,xN } ∈ RD , the objective of k-means is to partition X into K exhaustive and mutuall... |

891 | Least squares quantization in pcm
- LLOYD
- 1982
(Show Context)
Citation Context ... is the center of cluster Sk calculated as the mean (average) of points that belong to this cluster. This problem is known to be computationally intractable, but a heuristic method developed by Lloyd =-=[33]-=- offers a simple solution. Lloyd’s algorithm starts with K arbitrary centers, typically chosen uniformly at random from the data points. Each point is then assigned to the nearest center, and each cen... |

557 |
Hierarchical grouping to optimize an objective function
- Ward
- 1963
(Show Context)
Citation Context ...e initial centers is then recalculated as the mean of the points assigned to it. • Pairwise clustering (PWC) [12]: This method is a modification of Ward’s agglomerative hierarchical clustering method =-=[35]-=-. It builds a reduced color histogram and constructs a Q × Q joint quantization error matrix, where Q represents the number colors in the reduced color histogram. The clustering procedure starts with ... |

296 |
Feature discovery by competitive learning
- Rumelhart, Zipser, et al.
- 1986
(Show Context)
Citation Context ...) algorithm of Uchiyama and Arbib [1] is an online clustering algorithm based on the competitive learning paradigm, which is closely related to neural networks [32]. According to Rumelhart and Zipser =-=[34]-=-, a competitive learning scheme consists of the following three basic components: 1) Start with a set of units that are all the same except for some randomly distributed parameter which makes each of ... |

223 | Unsupervised segmentation of color-texture regions in images and video
- Deng, Manjunath
- 2001
(Show Context)
Citation Context ...play hardware has become more common, color quantization still maintains its practical value [2]. Modern applications of color quantization include: (i) image compression [3], (ii) image segmentation =-=[4]-=-, (iii) image analysis [5], and (iv) content-based image retrieval [6]. The process of color quantization is mainly comprised of two phases: palette design (the selection of a small set of colors that... |

194 |
Cluster analysis of multivariate data: Efficiency vs. interpretability of classifications
- Forgy
- 1965
(Show Context)
Citation Context ...on error are merged. In the experiments, the initial number of clusters was set to B = 20K. • Conventional k-means (KM): This method is implemented as outlined in Algo. (1) with random initialization =-=[36]-=- and a maximum number of iterations of 10. The effectiveness of a quantization method was quantified by the commonly used Mean Squared Error (MSE) measure: MSE = 1 HW H∑ W∑ h=1 w=1 ∥ ∥I(h, w) − ∥ Î(h,... |

143 | Color image quantization for frame buffer display
- Heckbert
- 1982
(Show Context)
Citation Context ...ring methods start with a single cluster that contains all N image pixels. This initial cluster is recursively subdivided until K clusters are obtained. Well-known divisive methods include median-cut =-=[3]-=-, octree [4], variance-based method [5], binary splitting [6], greedy orthogonal bipartitioning [7], center-cut [8], and rwm-cut [9]. On the other hand, agglomerative preclustering methods [10], [11],... |

81 | Color quantization of images
- Orchard, Bouman
- 1991
(Show Context)
Citation Context ...image pixels. This initial cluster is recursively subdivided until K clusters are obtained. Well-known divisive methods include median-cut [3], octree [4], variance-based method [5], binary splitting =-=[6]-=-, greedy orthogonal bipartitioning [7], center-cut [8], and rwm-cut [9]. On the other hand, agglomerative preclustering methods [10], [11], [12], [13] start with N singleton clusters each of which con... |

52 | An efficient color representation for image retrieval
- Deng, Manjunath, et al.
- 2001
(Show Context)
Citation Context ...ns its practical value [2]. Modern applications of color quantization include: (i) image compression [3], (ii) image segmentation [4], (iii) image analysis [5], and (iv) content-based image retrieval =-=[6]-=-. The process of color quantization is mainly comprised of two phases: palette design (the selection of a small set of colors that represents the original image colors) and pixel mapping (the assignme... |

34 |
Data clustering theory, algorithms, and applications
- Gan, Ma, et al.
- 2007
(Show Context)
Citation Context ...presents the experimental results, whereas Section 5 gives the conclusions. 2. Batch K-Means Algorithm The k-means algorithm [30] is inarguably one of the most widely used methods for data clustering =-=[31]-=-, [32]. Given a data set X = {x1, . . . ,xN } ∈ RD , the objective of k-means is to partition X into K exhaustive and mutually exclusive clusters S = {S1, . . .,Sk}, ⋃K k=1 Sk = X, Si ∩ Sj ≡ ∅ for i ̸... |

26 |
A Comparison of Clustering Algorithms Applied to Colour Image Quantization
- Scheunders
- 1997
(Show Context)
Citation Context ...tering methods at the expense of increased computational time. Clustering algorithms adapted to color quantization include kmeans [14], [15], [16], [17], minmax [18], [19], competitive learning [20], =-=[21]-=-, [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm [1] to the problem of color q... |

25 |
Variance-based color image quantization for frame buffer display
- Wan, Prusinkiewicz, et al.
- 1990
(Show Context)
Citation Context ...r that contains all N image pixels. This initial cluster is recursively subdivided until K clusters are obtained. Well-known divisive methods include median-cut [3], octree [4], variance-based method =-=[5]-=-, binary splitting [6], greedy orthogonal bipartitioning [7], center-cut [8], and rwm-cut [9]. On the other hand, agglomerative preclustering methods [10], [11], [12], [13] start with N singleton clus... |

25 |
A New Approach to Palette Selection for Color Images
- Balasubramanian, Allebach
- 1991
(Show Context)
Citation Context ...ian-cut [3], octree [4], variance-based method [5], binary splitting [6], greedy orthogonal bipartitioning [7], center-cut [8], and rwm-cut [9]. On the other hand, agglomerative preclustering methods =-=[10]-=-, [11], [12], [13] start with N singleton clusters each of which contains one image pixel. These clusters are repeatedly merged until K clusters remain. In contrast to preclustering methods that compu... |

19 |
Kohonen Neural Networks for Optimal Colour Quantization, Network Computation
- Dekker
- 1994
(Show Context)
Citation Context ...lgorithms adapted to color quantization include kmeans [14], [15], [16], [17], minmax [18], [19], competitive learning [20], [21], [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps =-=[26]-=-, [27], [28], [29]. In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm [1] to the problem of color quantization. This algorithm requires no cluster center initialization and a... |

19 | Adaptive color reduction
- Papamarkos, Atsalakis, et al.
- 2002
(Show Context)
Citation Context ...hms adapted to color quantization include kmeans [14], [15], [16], [17], minmax [18], [19], competitive learning [20], [21], [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], =-=[27]-=-, [28], [29]. In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm [1] to the problem of color quantization. This algorithm requires no cluster center initialization and avoids ... |

13 |
Color Image Quantization by Minimizing the Maximum Inter-cluster Distance
- Xiang
- 1997
(Show Context)
Citation Context ...uality results when compared to preclustering methods at the expense of increased computational time. Clustering algorithms adapted to color quantization include kmeans [14], [15], [16], [17], minmax =-=[18]-=-, [19], competitive learning [20], [21], [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama and Arbib’s competitive learning ... |

12 |
Color image quantization by agglomerative clustering
- Xiang, Joy
- 1994
(Show Context)
Citation Context ...t [3], octree [4], variance-based method [5], binary splitting [6], greedy orthogonal bipartitioning [7], center-cut [8], and rwm-cut [9]. On the other hand, agglomerative preclustering methods [10], =-=[11]-=-, [12], [13] start with N singleton clusters each of which contains one image pixel. These clusters are repeatedly merged until K clusters remain. In contrast to preclustering methods that compute the... |

11 | Color Image Quantization by Pairwise Clustering
- Velho, Gomes, et al.
- 1997
(Show Context)
Citation Context ... octree [4], variance-based method [5], binary splitting [6], greedy orthogonal bipartitioning [7], center-cut [8], and rwm-cut [9]. On the other hand, agglomerative preclustering methods [10], [11], =-=[12]-=-, [13] start with N singleton clusters each of which contains one image pixel. These clusters are repeatedly merged until K clusters remain. In contrast to preclustering methods that compute the palet... |

9 |
An adaptive clustering algorithm for color quantization
- Hsieh, Fan
- 2000
(Show Context)
Citation Context ... results when compared to preclustering methods at the expense of increased computational time. Clustering algorithms adapted to color quantization include kmeans [14], [15], [16], [17], minmax [18], =-=[19]-=-, competitive learning [20], [21], [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama and Arbib’s competitive learning algori... |

8 |
Center-Cut for Color Image Quantization
- Joy, Xiang
- 1993
(Show Context)
Citation Context ...ivided until K clusters are obtained. Well-known divisive methods include median-cut [3], octree [4], variance-based method [5], binary splitting [6], greedy orthogonal bipartitioning [7], center-cut =-=[8]-=-, and rwm-cut [9]. On the other hand, agglomerative preclustering methods [10], [11], [12], [13] start with N singleton clusters each of which contains one image pixel. These clusters are repeatedly m... |

6 |
A fuzzy algorithm for color quantization of images
- Özdemir, Akarun
- 2002
(Show Context)
Citation Context ...nse of increased computational time. Clustering algorithms adapted to color quantization include kmeans [14], [15], [16], [17], minmax [18], [19], competitive learning [20], [21], [22], fuzzy c-means =-=[23]-=-, [24], BIRCH [25], and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm [1] to the problem of color quantization. This algorith... |

5 |
Two high speed color quantization algorithms
- Brun, Mokhtari
(Show Context)
Citation Context ...e [4], variance-based method [5], binary splitting [6], greedy orthogonal bipartitioning [7], center-cut [8], and rwm-cut [9]. On the other hand, agglomerative preclustering methods [10], [11], [12], =-=[13]-=- start with N singleton clusters each of which contains one image pixel. These clusters are repeatedly merged until K clusters remain. In contrast to preclustering methods that compute the palette onl... |

4 |
Color Image Segmentation Using Rival Penalized Controlled Competitive Learning
- Law, Cheung
- 2002
(Show Context)
Citation Context ... methods at the expense of increased computational time. Clustering algorithms adapted to color quantization include kmeans [14], [15], [16], [17], minmax [18], [19], competitive learning [20], [21], =-=[22]-=-, fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm [1] to the problem of color quantiz... |

4 |
New adaptive color quantization method based on self-organizing maps
- Chang, Xu, et al.
- 2005
(Show Context)
Citation Context ...apted to color quantization include kmeans [14], [15], [16], [17], minmax [18], [19], competitive learning [20], [21], [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], [27], =-=[28]-=-, [29]. In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm [1] to the problem of color quantization. This algorithm requires no cluster center initialization and avoids the de... |

3 |
RWM-Cut for Color Image Quantization
- Yang, Lin
- 1996
(Show Context)
Citation Context ...usters are obtained. Well-known divisive methods include median-cut [3], octree [4], variance-based method [5], binary splitting [6], greedy orthogonal bipartitioning [7], center-cut [8], and rwm-cut =-=[9]-=-. On the other hand, agglomerative preclustering methods [10], [11], [12], [13] start with N singleton clusters each of which contains one image pixel. These clusters are repeatedly merged until K clu... |

3 |
Texture classification using nonlinear color quantization: Application to histopathological image analysis
- Sertel, Kong, et al.
(Show Context)
Citation Context ...ore common, color quantization still maintains its practical value [2]. Modern applications of color quantization include: (i) image compression [3], (ii) image segmentation [4], (iii) image analysis =-=[5]-=-, and (iv) content-based image retrieval [6]. The process of color quantization is mainly comprised of two phases: palette design (the selection of a small set of colors that represents the original i... |

2 |
An Algorithm for Competitive Learning in Clustering Problems
- Uchiyama, Arbib
- 1994
(Show Context)
Citation Context ...s algorithm. A common drawback of many conventional clustering algorithms is the generation of empty clusters (dead units). In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm =-=[1]-=- to the problem of color quantization. In contrast to the conventional batch k-means algorithm, this competitive learning algorithm requires no cluster center initialization. In addition, it effective... |

2 |
Digital Color Imaging Handbook
- Brun, Trémeau
(Show Context)
Citation Context ...splay hardware, which could not handle the 16 million possible colors in 24-bit images. Although 24-bit display hardware has become more common, color quantization still maintains its practical value =-=[2]-=-. Modern applications of color quantization include: (i) image compression, (ii) image segmentation, (iii) image analysis, (iv) image watermarking, and (v) content-based image retrieval. The process o... |

2 |
New Trends in Computer Graphics
- Gervautz, Purgathofer
- 1988
(Show Context)
Citation Context ... start with a single cluster that contains all N image pixels. This initial cluster is recursively subdivided until K clusters are obtained. Well-known divisive methods include median-cut [3], octree =-=[4]-=-, variance-based method [5], binary splitting [6], greedy orthogonal bipartitioning [7], center-cut [8], and rwm-cut [9]. On the other hand, agglomerative preclustering methods [10], [11], [12], [13] ... |

2 |
Graphics Gems Volume II
- Wu
- 1991
(Show Context)
Citation Context ...recursively subdivided until K clusters are obtained. Well-known divisive methods include median-cut [3], octree [4], variance-based method [5], binary splitting [6], greedy orthogonal bipartitioning =-=[7]-=-, center-cut [8], and rwm-cut [9]. On the other hand, agglomerative preclustering methods [10], [11], [12], [13] start with N singleton clusters each of which contains one image pixel. These clusters ... |

2 | A fast finite-state algorithm for generating RGB palettes of color quantized images
- Huang, Chang
- 2004
(Show Context)
Citation Context ... they can obtain higher quality results when compared to preclustering methods at the expense of increased computational time. Clustering algorithms adapted to color quantization include kmeans [14], =-=[15]-=-, [16], [17], minmax [18], [19], competitive learning [20], [21], [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama and Arbi... |

2 |
K-means Based Color Palette Design Scheme with the Use of Stable Flags, J Electron Imaging 16: 033003
- Hu, Lee
- 2007
(Show Context)
Citation Context ...can obtain higher quality results when compared to preclustering methods at the expense of increased computational time. Clustering algorithms adapted to color quantization include kmeans [14], [15], =-=[16]-=-, [17], minmax [18], [19], competitive learning [20], [21], [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama and Arbib’s co... |

2 |
Sample-size adaptive self-organization map for color images quantization, Pattern Recognit Lett 28
- Wang, Lee, et al.
- 2007
(Show Context)
Citation Context ...to color quantization include kmeans [14], [15], [16], [17], minmax [18], [19], competitive learning [20], [21], [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], [27], [28], =-=[29]-=-. In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm [1] to the problem of color quantization. This algorithm requires no cluster center initialization and avoids the dead uni... |

1 |
Color Quantization Using the Fast K-Means Algorithm
- Kasuga, Yamamoto, et al.
- 2000
(Show Context)
Citation Context ...ation, they can obtain higher quality results when compared to preclustering methods at the expense of increased computational time. Clustering algorithms adapted to color quantization include kmeans =-=[14]-=-, [15], [16], [17], minmax [18], [19], competitive learning [20], [21], [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama an... |

1 |
Accelerated K-means Clustering Algorithm for Colour Image Quantization
- Hu, Su
- 2008
(Show Context)
Citation Context ...tain higher quality results when compared to preclustering methods at the expense of increased computational time. Clustering algorithms adapted to color quantization include kmeans [14], [15], [16], =-=[17]-=-, minmax [18], [19], competitive learning [20], [21], [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama and Arbib’s competit... |

1 |
Local K-Means Algorithm for Colour Image Quantization
- Verevka, Buchanan
- 1995
(Show Context)
Citation Context ...reclustering methods at the expense of increased computational time. Clustering algorithms adapted to color quantization include kmeans [14], [15], [16], [17], minmax [18], [19], competitive learning =-=[20]-=-, [21], [22], fuzzy c-means [23], [24], BIRCH [25], and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm [1] to the problem of c... |

1 |
Fuzzy Clustering for Colour Reduction in Images
- Schaefer, Zhou
- 2009
(Show Context)
Citation Context ... increased computational time. Clustering algorithms adapted to color quantization include kmeans [14], [15], [16], [17], minmax [18], [19], competitive learning [20], [21], [22], fuzzy c-means [23], =-=[24]-=-, BIRCH [25], and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm [1] to the problem of color quantization. This algorithm requ... |

1 |
An Adjustable Algorithm for Color Quantization
- Bing, Junyi, et al.
- 2004
(Show Context)
Citation Context ...omputational time. Clustering algorithms adapted to color quantization include kmeans [14], [15], [16], [17], minmax [18], [19], competitive learning [20], [21], [22], fuzzy c-means [23], [24], BIRCH =-=[25]-=-, and self-organizing maps [26], [27], [28], [29]. In this paper, we apply Uchiyama and Arbib’s competitive learning algorithm [1] to the problem of color quantization. This algorithm requires no clus... |

1 |
Color Image Compression Using Quantization, Thresholding, and Edge Detection Techniques All Based on the Moment-Preserving Principle
- Yang, Tsai
- 1998
(Show Context)
Citation Context ...t images. Although 24-bit display hardware has become more common, color quantization still maintains its practical value [2]. Modern applications of color quantization include: (i) image compression =-=[3]-=-, (ii) image segmentation [4], (iii) image analysis [5], and (iv) content-based image retrieval [6]. The process of color quantization is mainly comprised of two phases: palette design (the selection ... |