Hierarchical SOM applied to image compression

Abstract

The increase of the need f o r image storage and transmission in computer systems has increased the importance of signal and image compression algorithms. The approach involving vector quantization (VQ) relies on designing of a j k i t e set of codes which will substitute the original signal during transmission with a minimal of distortion, , taking advantage of the spatial redundancy of image to compress them. Algorithms such as LGB and SOM work in an unsupervised way toward finding a good codebook for a given training data. However, the number of code vectors ( N ) needed f o r VQ increases with the vector dimension, and full-search algorithms such as LGB and SOM can lead to large training and coding times. An alternative for reducing the computational complexity is the use of a tree-structured vector quantization algorithm. This paper presents an application of a hierarchical SOM f o r image compression in which reduces rhe search complexity from O(N) to O(1og N), enabling a faster training and image coding. Results are given for conventional SOM, LBG and HSOM, showing the advantage of the proposed method,

SEAN: A simple expanding-tree algorithm based on mean-division for color quantization

“…Some well-known splitting algorithms are the median-cut [13], center-cut [14], Octree [15], variance-based [16], RWM-cut [17], binary splitting [19]. Clustering-based methods include K-means [8], [10], [11], [12], fuzzy c-means [20], [21] and self-organizing maps [9], [22], [23], [24], [25].…”

LazySOM: Image Compression Using an Enhanced Self-Organizing Map

“…For fair comparison, the parameters of HSOM approach were set as in paper [8]. Moreover, the stopping threshold (ε) of LBG was set to 0.0001, while the training epoch was set to 200 (for conventional SOM).…”

“…However, SOM takes a full search to compete with the training data set, so has a fairly high time complexity. Many elevated SOMs, such as the standard and advanced Fast SOM systems, try to decrease the large computation effort of SOM [8], [9]. These schemes focus mainly on the initialization of the neural network.…”

“…VQ is a lossy compression approach [1], [2]. Correlative researchers have developed several vector quantization methods, such as LBG [3] , SOM [4]- [7] and HSOM [8], [11], the most popular being SOM. Since SOM is a full search technique, it can find an impressive codebook for image reconstruction, but incurs a large computation effort, and has a time complexity of O(n).…”

Diseño y experimentación de un cuantizador vectorial hardware basado en redes neuronales para un sistema de codificación de video.

“…En [83] se desarrolla y experimenta una red SOM jerárquica (HSOM) que reduce la complejidad temporal de O(N) a O(logN) para disminuir el tiempo de entrenamiento. Los autores emplean la técnica de cuantización vectorial estructurada enárbol para generar la red HSOM, quedandoésta conformada como un conjunto de redes SOM organizadas en unárbol jerárquico.…”