Automatic pigment network classification using a combination of classic texture descriptors and CNN features
The presence of atypical (irregular) pigment networks can be a symptom of melanoma malignum in skin lesions, thus, their proper recognition is a critical task. For object classification problems, the application of deep convolutional neural nets (CNN) receives priority attention in these days for their high recognition rate. The descriptive features found by CNNs are more hidden than the classically applied ones for texture recognition. In this paper, we investigate whether CNN features outperform the classic texture descriptors in the classification of typical/atypical pigment network. Beyond performing this analysis, we have also found that the aggregation of CNN and classic features within a joint classification framework had a superior performance. Especially, the mixed feature set leads to a much higher stability in classification performance for various classifiers. As for quantitative figures, we have reached 90.44% recognition accuracy using the mixed features and a Random Forest classifier boosted with linear forward feature selection.