Research on CAD based boundary-feature descriptions of skin lesions for early diagnosis and medical treatments are crucial as boundary irregularity of melanomas is one of the most important clinical indicators to discriminate lesiens from other benign moles. Conventional fractal dimension（FD） is utilized to describe boundary complexity based on self-similarity measures. The criminative power of FDs is low due to its global property and lack of detail information. The objective of the paper is to explore new irregularity descriptors of object boundaries at different scales. In the paper

there are two novel measures: the irregularities of boundary structures and multi-scale features related to boundary roughness are described using an integrated quantitative model combining Gaussian filtering and local FDs. The advantage of the latter approach is that feature clusters of boundary irregularity are formed at different scales in the image space

so that the extracted measurements from multi-scale descriptions and different viewpoints（statistics and boundary-like roughness） can provide effective descriptors compared to the conventional FD and other boundary based methods and will be helpful for further classification tasks. Experiments show that the proposed irregular descriptors not only enhance fine expressions of boundary complexity but also can effectively discriminate melanomas among moles using statistical features of the boundary roughness at different scales.