Image processing with learning capabilities for color images

Gulzhan Zhassulanbaikyzy (1)

1 EDP - Equations aux Dérivées Partielles

The application of anisotropic diffusion equation for color image processing is discussed in this work. Nonlinear PDEs of diffusion type are a common tool for color image processing and image filtering. These PDEs are designed in such a way that diffusion eliminates the noise but preserves the significant edges and features of the image. A diffusion tensor with learning abilities was designed in to provide these features. This model was previously applied to grey-scale images and gave excellent results [9]. In this work this model is extended to color images. The proposed algorithm for color image denoising is implemented in the form of C++ scripts. In order to evaluate our proposed algorithm, comparisons with Chambolles’s projection algorithm [21, 6] and the more recent DA3D algorithm [38] are performed. The source code of both algorithms the description of the methods and the test images were borrowed from the IPOL online resources. The results confirm the good performance of the diffusion tensor approach for color images.

Domaines: Analyse numérique [math.NA]

Gulzhan Zhassulanbaikyzy. Image processing with learning capabilities for color images. [Research Report] Laboratoire Jean Kuntzmann; Université Grenoble - Alpes; Grenoble INP. 2017. ⟨hal-01576540⟩