In-Depth Analysis of Automated Acne Disease Recognition and Classification

TL;DR

The paper tackles automated recognition and six-class classification of facial acne to aid dermatologists. It presents a pipeline that preprocesses images (contrast enhancement, smoothing, LAB color space), segments acne regions with k-means, and extracts texture features by combining GLCM and statistical descriptors before classifying with five learners (LR, DT, KNN, SVM, RF). The Random Forest classifier achieves the best performance, with an average accuracy of 98.50% and strong precision, sensitivity, and specificity, along with an AUC of 85.8%. The method demonstrates superiority over AcneNet baselines on public and larger six-class datasets, and includes analysis of feature contributions and misclassification patterns. The work highlights practical potential for automated screening, though limitations include misclassifications under severe or blurry imaging and the need for more uniform datasets to support robust grading.

Abstract

Facial acne is a common disease, especially among adolescents, negatively affecting both physically and psychologically. Classifying acne is vital to providing the appropriate treatment. Traditional visual inspection or expert scanning is time-consuming and difficult to differentiate acne types. This paper introduces an automated expert system for acne recognition and classification. The proposed method employs a machine learning-based technique to classify and evaluate six types of acne diseases to facilitate the diagnosis of dermatologists. The pre-processing phase includes contrast improvement, smoothing filter, and RGB to L*a*b color conversion to eliminate noise and improve the classification accuracy. Then, a clustering-based segmentation method, k-means clustering, is applied for segmenting the disease-affected regions that pass through the feature extraction step. Characteristics of these disease-affected regions are extracted based on a combination of gray-level co-occurrence matrix (GLCM) and Statistical features. Finally, five different machine learning classifiers are employed to classify acne diseases. Experimental results show that the Random Forest (RF) achieves the highest accuracy of 98.50%, which is promising compared to the state-of-the-art methods.

Figures (8)

• Figure 1: An overview of the proposed expert system for recognizing and classifying acne disease. Input images are pre-processed through contrast enhancement, smoothing filter, and L*a*b color conversion and then segmented using k-means clustering. The GLCM matrix and Statistical feature extraction methods are employed to extract features. The five classifiers, namely logistic regression, decision tree, K-nearest neighbors, random forest, and support vector machine, are utilized to classify acne diseases.
• Figure 2: Visualization of the L*a*b color conversion process step by step. Here, (a), (b), (c), and (d) denote the visualization of L, a, b, and L*a*b conversions, respectively.
• Figure 3: Segmentation and feature extraction results of acne disease are taken as examples. The following list including, (a), (b), (c), (d), (e), and (f) are extracted feature matrics ($C$, $\rho$, $E$, $S$, $H$, $\mu$, $\sigma$, $\sigma^2$, $K$, RMS, Smoothness, Skewness, and cluster shade (Cs)) of ACC, AC, AE, AK, AOC, and AP, respectively.
• Figure 4: Multi-class confusion matrix for the RF classifier.
• Figure 5: AUC-ROC curves of five different ML classifiers.