Classification of Indonesian Batik: A Comparative Study of CNN and VGG16 with Canny Edge Detection

Batik is a traditional Indonesian art that uses wax-resist dyeing to create intricate patterns known as batik motifs on fabric. These motifs, influenced by local culture, encompass various types, including Banyumasan batik, which can be challenging to identify due to its derivation from main patterns. Convolutional Neural Networks (CNN) is a robust method for recognizing patterns in the image domain. CNN can also perform pattern recognition tasks like image classification and segmentation. The objective of this study is to assess the effectiveness of using Canny edge detection combining convolutional neural network (CNN) and VGG16 for classifying Batik Banyumasan based on images. We developed a primary dataset collected from the House of Batik R which is a reputable institution in the Banyumas district of central Java, Indonesia. It is known for its extensive collection of batik motifs. The dataset consists of 5148 images classified into eleven classes. In the image processing stage, we apply Canny edge detection to determine the boundaries of the motif image. In our experiment, we developed four distinct CNN models (M1, M2, M3, and M4) and compared to pre-trained VGG16. To improve the accuracy and mitigate overfitting, the experiment applied specific hyperparameters to these models. In conclusion, VGG16 performs better than the basic CNN. The use of Canny edge detection actually reduces the model’s performance although it improves generalization. These findings suggest that Canny edge detection is not effective in recognizing the complex features of Batik image. Using Canny in combination with a CNN architecture is not an optimal choice.