COMPARATIVE ANALYSIS OF CONTRAST ENHANCEMENT METHODS FOR CLASSIFICATION OF PEKALONGAN BATIK MOTIFS USING CONVOLUTIONAL NEURAL NETWORK

Batik artists in Pekalongan have freedom in determining motifs, creating a diversity of distinctive batik motifs. However, this diversity often makes it difficult for people to recognize the different motifs, as visual identification requires in-depth knowledge. The lack of understanding about Pekalongan batik is a challenge in recognizing these motifs. To overcome this challenge, an efficient and accurate method of motif identification is needed. This study aims to analyze the efficacy of contrast enhancement methods in improving the classification results of Pekalongan batik motifs using convolutional neural networks (CNN) with ResNet50 architecture. The dataset of 480 images was collected directly from Museum Batik Pekalongan and split into three distinct categories: 15% for validation, 15% for testing, and 70% for training. Two contrast enhancement methods: contrast limited adaptive histogram equalization (CLAHE) and histogram equalization (HE), were applied to create additional datasets. The Adam optimizer was used to train the model over 50 epochs at a learning rate of 0.001. The test results show that the original dataset contrast-enhanced with CLAHE reaches the best accuracy of 83%, followed by the original dataset contrast-enhanced with HE at 81%, and the original dataset at 76%. This finding shows that the application of contrast enhancement methods, especially CLAHE, can increase the model's accuracy in classifying batik motifs.