Quality and Randomness Assessment of Images Generated by a Convolutional Autoencoder

This paper presents an analysis of the quality and randomness of information in images generated with a convolutional autoencoder (CAE). The CAE convolved altered color images from CIFAR-10 dataset. The CIFAR-10 images were altered by randomly setting 30%, 60%, and 90% of pixel values to [0, 0, 0] or [255, 255, 255], respectively. In the validation stage, Mean Square Error (MSE) loss function reached 0.0115 and the accuracy metric 0.7461. Similarity Structural Index Measure (SSIM), Pearson Correlation Coefficient (PCC) and Peak Signal-to-Noise Ratio (PSNR) metrics, assessed quality of generated images. The assessment results ranged as follows: SSIM [0.3251, 0.6830], PCC [0.5034, 0.9358], and PSNR [18.63 dB, 26.04 dB]. The Shannon metric assessed randomness both locally and globally for each image, ranging from 1.25 to 2.49 bits, and from 6.61 to 9.71 bits, respectively. CAE implementations highlight their potential for applications in technological innovation. Smart citations: https://scite.ai/reports/10.61467/2007.1558.2026.v17i2.952Dimensions.Open Alex.