A convolutional autoencoder and a neural gas model based on Bregman divergences for hierarchical color quantization

• Published in: Neurocomputing (Amsterdam) 2023-08, Vol.544, p.126288, Article 126288
• Main Authors: Fernández-Rodríguez, José David, Palomo, Esteban J., Benito-Picazo, Jesús, Domínguez, Enrique, López-Rubio, Ezequiel, Ortega-Zamorano, Francisco
• Format: Article
• Language: English
• Subjects: clustering; color quantization; convolutional autoencoder; self-organization
• ISSN: 0925-2312; 1872-8286
• DOI: 10.1016/j.neucom.2023.126288

•We present a novel hybrid model composed of a convolutional autoencoder and a Growing Hierarchical Bregman Neural Gas (GHBNG) for hierarchical color quantization.•The performance of our proposed hybrid model attains better compressed sizes than the autoencoder alone.•In the cases that autoencoders alone yield about the same compression ratio, our hybrid model attains a better compressed image quality as measured by the PSNR and SSIM performance metrics.•Our proposal learns a reduced representation of an input image with a tunable compression strength given by the number of convolutional layers of the encoder section of the autoencoder (N parameter).•GHBNG gets better compression ratios when compared with other hierarchical selforganizing models. Color quantization (CQ) is one of the most common and important procedures to be performed on digital images. In this paper, a new approach to hierarchical color quantization is described, presenting a novel neural network architecture integrated by a convolutional autoencoder and a Growing Hierarchical Bregman Neural Gas (GHBNG). GHBNG is a CQ algorithm that allows the compression of an image by choosing a reduced set of the most representative colors to generate a high-quality reproduction of the original image. In the technique proposed here, an autoencoder is used to translate the image into a latent representation with higher per-pixel dimensionality but reduced resolution, and GHBNG is then used to quantize it. Experimental results confirm the performance of this technique and its suitability for tasks related to color quantization.