Texture analysis for dermoscopic image processing

A new method for detecting pigment network which is one of the textures often visible in skin lesions is presented. In dermatoscopy, this texture is used as one of the features used to evaluate likeliness of cancer and can indicate if a lesion is of malignant nature. Method presented here uses an adaptive filter inspired by Swarm Intelligence (SI) optimization algorithms. The filtering method introduced here is applied to dermoscopic skin image in a non-linear manner and allows selective image filtering. First stage of filtration process is to randomly spread the agents (swarm member) throughout the two-dimensional space (processed image), where each of those agents adapts its parameters to best fit the local neighborhood. In next steps of filtration process, the agents can share information with other swarm members that are located in immediate vicinity. This approach is new to the problem of dermoscopic texture detection, and is highly flexible, as it can be applied to images without the need of previous pre-processing. This feature is highly desirable due to the fact that in most cases of computer aided diagnostic, input images need to be pre-processed (e.g. for brightness normalization, histogram equalization, contrast enhancement, color normalization) and this can results in unwanted artifacts or simply may require human verification. Introduced method was developed specially to recognize one of the differential structures (pigmented network texture) used for calculating the Total Dermoscopy Score (TDS) of the ABCD rule