Batik Style Transfer into Fractal Shape

Image style transfer is a technique that blends the features of two images. The application of neural style transfer extends to Indonesian batik. Previous studies worked on creating new batik patterns by transferring the batik style to another image. The goal of this paper is to generate batik-like motifs by forming fractal shapes colored with migrated batik style. This was done by having fractal acts as the content image and batik as the style image. Unlike previous works, an evaluation procedure was performed in this study to assess the result of the applied style transfer model. The assessment was done to both methods of previous studies and the proposed method from this paper. VGG19 architecture was chosen to act as the style transfer model for this study. Though this network is one of the oldest architectures, it has the best convolutional layer to extract features of batik images. The architecture was equipped with different loss functions. Total variation was added to the function to help denoise the quality of the generated image. The generated image was also given an image-filtering procedure to capture the formation of the fractal. Some questionnaires were used to help evaluate the results of this study. The filtered image with the proposed method yielded slightly more positive opinions than the generated image without further filtering. However, the hypothesis tests showed that the different treatments did not produce significant results. The questionnaires also showed style transfer result was easily identifiable as not man-made, distinguishing it from traditional batik.