**4. Conclusions**

The proposed method for detection and classification of burns on images is found to be adequate as an assisting tool in the diagnosis of skin burns and it offers nonintrusiveness and low economic cost. The tools and analyses in the current work contribute to skin burn detection and classification: (1) Achieving high classification performance (sensitivity and precision above 90\%); (2) sparse reconstruction of image patches with dictionaries was shown to be adequate for this application; (3) Analyzing images under diverse conditions of skin color (ethnicity) and illumination. It was found that untrained dictionaries outperformed trained dictionaries for this application. The two groups of features most commonly used in the imaging-based classification of burns, namely color and texture, provided the reported method with good classification performance.
