Tamayo Zapata, Kelly J.

Profile Picture

Filters

20241
Reset filters

Settings

Citations

Publication Search Results

Now showing 1 - 1 of 1
  • Thumbnail Image
    Publication
    Quality assessment of images from projection devices using deep neural networks
    (2024-05-08) Tamayo Zapata, Kelly J.; Manian, Vidya; College of Engineering; Ducoudray, Gladys; Torres, Raul E.; Department of Electrical and Computer Engineering; Font Santiago, Cristopher B.
    Image quality perception in projectors is critical for enhancing user experience in education, business, and entertainment. This research proposes an automated visual assessment tool for accurate image quality assessment (IQA) in DLP projector images. The research utilizes two types of neural networks: No-Reference IQA (NR-IQA) for evaluating blur, blob, and pixel defects, and Full-Reference IQA (FR-IQA) for evaluating color distortion. The methodology involves neural network models, transitioning from a pre-trained VGG16 to a more complex ResNet, and ultimately a refined ResNet + PyramidNet for NR-IQA for the NR-IQA; and a Simasese pre trained VGG16, and Siamese ResNet models for FR-IQA. For the refined NR-IQA ResNet + PyramidNet model AUC of 94%, 96%, 98, 94% in high-quality, blur, blob, and pixel defects, respectively were achieved, with good test outcomes, particularly in projectors with visual defects. Conversely, the FR-IQA Siamese ResNet model achieved AUC rates of 94% and 96% for high quality and color-altered images, respectively. It also demonstrated good performance with real defective scenarios. Limitations were encountered in both situations, where intentionally unfocused images or the presence of shadows caused misclassification of blur or blobs, or when handling new images captured under varying conditions. Despite the limitations, incorporating a diverse dataset, including both simulated and real defective projector images, significantly enhanced model performance and defect identification accuracy.