**1. Introduction**

The human body is covered by skin and this muscle protects all the others. This muscle is essential for the human senses. It not only provides sensation but also helps with the skin regulation and synthesis of vitamin D. Skin can get damage by burn accidents. In a burn injury, some skin layers are destroyed, and first treatments are required as soon as possible. These treatments are based on intensity and severity. Burn area, depth, and location are the determining factors for the gravity. While partial-thickness burns can heal immediately with minimal scarring, deep partialthickness and full-thickness burns require more than three weeks to close. They are often associated with significant scarring and functional limitations unless removed and grafted within the first days of the injury [1]. Thus, the correct diagnosis of burn depth is necessary to achieve optimal results. It is also known that burn injury may extend to grow over the first days heading to the conversion of superficial burns to deep burns. A better perception of the mechanisms leading to burn wound regeneration is probable to lead to new treatments that can result in the limitation of burn

wound progression, which leads to better healing. Therefore, it is vital to identify the type of skin and the position of the injury [2]. Despite its importance, verified methods to measure wound closure are lacking, making any comparisons of innovative therapies difficult. Wound healing and long-term consequences are determined by burn depth. The distinction between each degree lies in how many layers of skin are damaged [3].

Research interest in systems for automatic diagnosis of skin burns is motivated to assist health care personnel in certain situations. Clinics in small towns may rely on the work of interns or professional apprentices. Additionally, quantitative analysis of important burnt skin features can be used as contextual information for clinical interpretation, for recommendation of treatment. Classification of skin burns is also useful in assisting healing since the treatment depends on the burn degree and size.

Previous work in this area has focused on the classification of burn wounds by applying various artificial vision approaches. Cognitive systems are machine models whose function is inspired by the way that the human brain works. The use of cognitive systems is not intended to replace humans but rather to augment human capabilities through automatic assistance tools [4]. Although there is much research done, there are many opportunities in the area. Segmentation and image processing are fields widely used for artificial vision and medical imaging, especially for burn detection applications [5–9].

Şevik et al. [10] proposed a model to classify burns with texture-based features to identify between burnt skin, healthy skin, and background. The authors combined various algorithms to obtain a model that can find burn wounds with a small error. They also developed different models to compare them all and choose the best results as the most accurate method. The goal was to classify the images into three different objects: skin, burn, and background regions. The best process was by applying the fuzzy c-means algorithm at the stage of the segmentation. In the next step of the algorithm, a multilayer feed-forward artificial neural network trained with the backpropagation algorithm was used for classification. The authors created their database in collaboration with Turkish hospitals. It contains 105 images from various patients. The photos had to be pre-processed to standardize the size of the pictures. The authors used texture patterns from Haralick features from the Gray Level Co-occurrence Matrix (GLCM) for the feature extraction process. From a different perspective, Rehman et al. [11] proposed a computer-aided diagnosis (CAD) for the classification of the burnt skin. Their focus lies at the identification of the depth of the injury. Their objective was to identify the thickness of injured human skin and then be able to classify among first, second, and third-degree burns. The authors used the Otsu method of thresholding for the segmentation step and then applied the statistical method GLCM to obtain the feature vector for the input of the classifier. They used contrast and correlation instead of texture. The author created their database with the help of the Burn Center of Allied Hospital Faisalabad, Pakistan. It contains around 173 images from the different types of burn depth. Yadav et al. [12] suggested a classification model to diagnose burns based on support vector machine (SVM). The training is performed by classifying images into two classes, those that need grafts and those that are non-graft. They skipped the segmentation step and converted the pictures to CIELAB space since the range of colors on that scale is more extensive than RGB. They obtained Hue, Kurtosis, Skewness, and Chroma to determine the area and depth of the wound and, based on the analysis, classify the injury in graft or non-graft with the help of the SVM.

*Detection and Classification of Burnt Skin on Images with Sparse Representation of Image… DOI: http://dx.doi.org/10.5772/intechopen.105162*

The rest of the manuscript is organized as follows. The following section describes the methodology for feature extraction and the image segmentation and classification methods. The detection and classification results are reported in the Results section. Finally, conclusions are presented in the last section.
