Shading and texture constrained retinex for correcting vignetting on dermatological macro images

• Published in: The Visual computer Vol. 39; no. 2; pp. 693 - 709
• Main Authors: Sathish, S., Sumithra, M. G., Mohanasundaram, K.
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.02.2023; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Automation; Color; Computer Graphics; Computer Science; Computing time; Constraints; Convex analysis; Convexity; Cost function; Decomposition; Dermatology; Digital cameras; Digital imaging; Illumination; Image Processing and Computer Vision; Image reconstruction; Melanoma; Microscopy; Optimization; Original Article; Photography; Saturation (color); Shading; Texture; Vignetting; Retinex decomposition; Image processing; Vignetting error; Illumination-correction; Dermatological photography
• ISSN: 0178-2789; 1432-2315
• DOI: 10.1007/s00371-021-02368-z

Because of the flexibility and availability of high-resolution digital cameras, dermatological photography is considered as a good alternative to dermoscopy. However, uneven background illumination on the dermatological photographs makes their automated analysis troublesome. Equalization of the uneven background illumination is helpful to make the automated analysis of the dermatological photographs more sensitive and specific. A customized algorithm for equalizing the uneven background illumination on dermatological photographs is proposed in this paper. The illumination-corrected image is reconstructed from the gamma-corrected illumination component in Hue Value Saturation (HSV) color space. The Retinex decomposition of the value component is formulated as a non-convex optimization problem. Constraints within the cost function are derived from the shading and texture priors. The shading and texture priors are computed respectively from the derivatives of the illumination and texture priors. On 137 dermatological photographs, the values of Average Gradient of the Illumination Component, Lightness Order Error, Sparse Feature Fidelity, Visual Saliency-based Index, Visual Information Fidelity and the computational time exhibited by the proposed devignetting algorithm are 0.1895 ± 0.0386, 232.9553 ± 140.7912, 0.9783 ± 0.0106, 0.9903 ± 0.0021, 0.7063 ± 0.0396 and 2.0272 ± 0.4319 (sec). The proposed algorithm is able to equalize the uneven background illumination without scaling or boosting it intolerably. It produces output images that are natural in appearance and free from structural/color artefacts. The loss of salient information is negligible in the proposed algorithm. It is computationally fast, as well.