Systems, Signals and Image Processing by Unknown

Author:Unknown
Language: eng
Format: epub
ISBN: 9783030968786
Publisher: Springer International Publishing

(2)

where and are two small positive constants, , , . The SSIM index value 1 is reached if and only if x and y are identical.

Fig. 2.Example of similar images and the MSE and SSIM values between them.

We discarded and replaced all images with MSE index below 3000, and SSMIM index above 8.5. Figure 2 presents an example of two images of the same class, and the similarity degree between them using the MSE and SSIM indexes.

3.3 Data Augmentation

Due to the small amount of images in the database, we used the data augmentation technique to overcome this problem and avoid over-fitting.

The problem of over-fitting reduces the ability of CNNs to generalize unsupervised data. Thus, the data augmentation process was necessary to enrich the training set, adding new samples in their respective classes, with transformations such as scale, zoom and translations [20]. Ten new images were generated for each original image from each class, totaling a total set of 9900 images. Figure 3 shows transformations of zoom range, width shift range and height shift of 0.3, horizontal flip, vertical flip and 90 rotation range performed during the data augmentation process, which may occur simultaneously.

Fig. 3.Examples of data augmentation.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.