A Colour Palette for Automatic Detection of Blue-White Veil

Colour assessment of pigmented skin lesions are essential for the diagnosis of malignant melanoma. However, visual interpretation of colour is subjective and prone to error. Computer programs can provide support to clinicians to overcome this subjectivity. So far, methods for colour analysis of this nature have utilised statistical classification models. This paper puts forward an alternative framework: an effort to reproduce the experience of human observer. The proposed method introduces a perceptually intuitive and semantically meaningful approach for colour and colour-related feature detection. As a case study, the task of automatic detection and segmentation of blue-white veil feature in dermoscopy images is examined. Our proposed method, as shown in our experiments, outperforms the prior art for this task, while it attempts to mimic the human perception of skin lesion colours.