Image compression using hybrid neural networks combining the auto-associative multi-layer perceptron and the self-organizing feature map

Abidi, Mongi A.; Yasuki, S.; Crilly, P.B.

doi:10.1109/30.338325

Cited by 9 publications

(2 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some examples of these techniques are Multidimensional Scaling (MDS), Locally Linear Embedding (LLE), Isomap, Kernel PCA [7], Self-Organizing Maps (SOM) [8] [9] [10] and also Auto-Associative Neural Networks (AANN) [1] [2]. The latter, AANN or also known as Bottleneck Neural Networks (BNN) has been previously used for data compression or dimension reduction particularly in the field of information retrieval, chemical applications, missing data estimations and image compressions [2] [6] [11] [12]. But, there was still very little attention given to AANN.…”

Section: Introductionmentioning

confidence: 99%

Dimension Reduction and Clustering of High Dimensional Data using Auto-Associative Neural Networks

Zin

Yusof

Ehsan

2013

IJCA

View full text Add to dashboard Cite

The task to capture and interpret information hidden inside high-dimensional data can be considered very complicated and challenging. Usually, dimension reduction technique may be considered as the first step to data analysis and exploration. The focus of this paper is on high-dimensional data dimension reduction using a supervised artificial neural networks technique known as Auto-Associative Neural Networks (AANN). The AANN can be considered as a powerful tool in data analysis and clustering with the ability to deal with linear and nonlinear correlation among variables. This technique is sometimes referred to as nonlinear principal component analysis (NLPCA), Encoding-Decoding networks, or bottleneck neural networks (BNN) due to its unique structure. It reduces high-dimensional data into low-dimensional data on its bottleneck layer which can later be used for data transmission, clustering and visualization. In this paper, a structurally flexible AANN is developed by using high level computer language, applied and studied on two case studies of Iris flowers and Italian olive oils datasets. The purpose of the work was to investigate the ability of AANN to reduce dimension of high-dimensional data on small (Iris) and large (Olive) datasets. The results have shown that AANN has been able to compress high-dimensional data into only one or two non-linear principal components at its bottleneck layer with the highest accuracy of 98.9% and 82.1% for both datasets respectively. AANN has also managed to perform accurately in both reducing dimension and clustering data by only using small portion of training dataset.

show abstract

Section: Introductionmentioning

confidence: 99%

Dimension Reduction and Clustering of High Dimensional Data using Auto-Associative Neural Networks

Zin

Yusof

Ehsan

2013

IJCA

View full text Add to dashboard Cite

show abstract

“…Three layer perceptron networks with an identical number ofinput and output units and fewer hidden units are found to be able to perform principal component analysis when trained by the backpropagation algorithm [8] to duplicate the input pattern at the output, and have been successfully applied to image compression [19]. Researchers have combined different neural network models together [20,21] Neural network approaches to image compression are block based coding techniques. As with other block based image compression techniques, these methods cannot avoid the problem of annoying "blocking effects" in the reconstructed images.…”

mentioning

confidence: 99%

<title>Finite-state neural networks for dimensionality reduction and smooth signal reconstruction</title>

Qiu

Varley

1996

Applications of Artificial Neural Networks in Image Processing

View full text Add to dashboard Cite

A finite-state auto-associative MLP neural network is studied in the context of dimensionality reduction and smooth signal reconstruction. We describe the structure and the training procedure of the finite-state network. One of the desirable properties ofthe auto-associative MLP is that the variance ofthe hidden Units' OUtputs can be arranged in a descending order, so that efficient coding of the hidden layer output can be implemented. We provide experimental results to demonstrate that the finite-state network retains this desirable property of its memory-less counterpart. One of the application areas of the auto-associative MLP is image compression. As with other block based image compression techniques, this method cannot avoid the problem of annoying "blocking effects" in the reconstructed images. We present simulation results to demonstrate that the finite-state auto-associative MLP can be used to achieve effective image data compression whilst significantly reducing the blocking effects.

show abstract

Image Compression Based on Genetic Algorithm and Deep Neural Network

Deng

Liu

Wang

et al. 2016

Bio-Inspired Computing – Theories and Applications

View full text Add to dashboard Cite

Image compression using hybrid neural networks combining the auto-associative multi-layer perceptron and the self-organizing feature map

Cited by 9 publications

References 17 publications

Dimension Reduction and Clustering of High Dimensional Data using Auto-Associative Neural Networks

Dimension Reduction and Clustering of High Dimensional Data using Auto-Associative Neural Networks

<title>Finite-state neural networks for dimensionality reduction and smooth signal reconstruction</title>

Image Compression Based on Genetic Algorithm and Deep Neural Network

Contact Info

Product

Resources

About