Influence of Image Pre-processing to Improve the Accuracy in a Convolutional Neural Network
Convolutional neural networks (CNN) have been applied in different fields including image recognition. A CNN requires a set of images that will be used to teach to classify it into specific categories. However, the question about how image pre-processing influences CNN accuracy has not yet been answ...
Saved in:
| Other Authors: | , , , |
|---|---|
| Format: | Artículo |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | https://ijcopi.org/index.php/ojs/article/view/163 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Convolutional neural networks (CNN) have been applied in different fields including image recognition. A CNN requires a set of images that will be used to teach to classify it into specific categories. However, the question about how image pre-processing influences CNN accuracy has not yet been answered bluntly. This paper proposes the application of pre-processing methods for the images’ feed to a CNN in order to improve the accuracy of the classification. Two methods of pre-processing are evaluated, quantization and sharpness enhancement. Quantization carries out at 7 levels, and sharpness works with four levels using the discrete wavelet transform. The tests were implemented with two CNN models, LeNet-5 and ResNet-50. In the first part of this paper the methodology and description of the CNN models, as well as the data set are presented. Later the descriptions of the experiments are presented. Finally, it is shown how the proposed method achieved an improvement on the accuracy compared with the results obtained with the images with no modification. The proposed pre-processing methods had an improvement between 1.35 and 3.1% on the validation accuracy. |
|---|