New pruning techniques for constructive neural networks with application to image compression

Liying Ma; Khashayar Khorasani

doi:10.1117/12.395080

4 August 2000 New pruning techniques for constructive neural networks with application to image compression

Liying Ma, Khashayar Khorasani

Proceedings Volume 4052, Signal Processing, Sensor Fusion, and Target Recognition IX; (2000) https://doi.org/10.1117/12.395080
Event: AeroSense 2000, 2000, Orlando, FL, United States

Abstract

Image compression is an important research domain in image processing. Recently, several neural netowkr (NN) based schemes developed in this are. In particular, constructive feed-forward neural networks have been attempted by many researchers to this problem. The constructive NN-based schemes are promising given their lower training cost, satisfactory performance and automatic determination of proper network size. In this paper, we first consider a NN- based technique that uses a constructive one-hidden-layer FNN for image compression. In standard NN-based schemes when a new hidden unit is added to the net the whole net is retrained while in this scheme the input-side weights are first trained and then all the network output-side weights are adjusted, resulting in a considerably less computational efforts. Next, two pruning techniques are proposed to remove the unnecessary input-side weights during the network construction, without sacrificing the performance of the network, to yield a smaller and a more economical network. To confirm the effectiveness of the prosed techniques, we have applied them to both regression problems and image compression. It has been found that a significant number of weights can be pruned without degenerating the network performance.

Citation Download Citation

Liying Ma and Khashayar Khorasani "New pruning techniques for constructive neural networks with application to image compression", Proc. SPIE 4052, Signal Processing, Sensor Fusion, and Target Recognition IX, (4 August 2000); https://doi.org/10.1117/12.395080

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available