Image Enhancement using Hybrid Convolutional Autoencoder with Clahe Post Processing
Pranika, Amandeep · International Journal of Enhanced Research in Management & Computer Applications · 2025
In computer vision, improving low-resolution images is necessary when compression, sensor fails or outside conditions result in image degradation. The author introduces a hybrid technique that uses a Convolutional Autoencoder (CAE) and Contrast Limited Adaptive Histogram Equalization (CLAHE) to enhance both image sharpness and local contrast. The architecture of the CAE is set up to map small, pixelated images into good reconstructions using convolutional layers with sigmoid code, max pooling, upsampling and adding shortcuts that help keep important low-level features. For this task, input photos are made grainy which allows the network to remember and restore the original quality. When the structural content is restored by CAE, CLAHE is used to optimize contrast in different parts of the image, reduce disturbances and give the image greater clarity. Real-world RGB images of size 80×80 were used to train the model with Adam and MSE and early stopping and checkpointing were implemented to maintain a good performance.