Reliable Batik Image Classification : Mulwin-LBP Algorithm and Deep Neural Network

This study proposed feature extraction using the Mulwin-LBP algorithm which can optimally supported batik image classification activities. In this classification activity, the Deep Neural Network algorithm was used to provide solutions to the Invariant dilemma. However, previously, the image segmentation process using the Watershed method would be carried out to support the feature extraction. Actually this feature extraction algorithm is done by combining 3 and 4 different types of windows. The window sizes were used 6x6 pixel, 9x9 pixel, 12x12 pixel, and 15x15 pixel or a combination of 3 or 4 windows to get optimal and maximum results. Some experiments included the Training Images number more than the test image number, but the accuracy and precision of classification could reach more than 72.88%. Meanwhile, if the test was carried out by adding image classes, including the number of training images more than test images, the resulting classification accuracy had reached more than 81.65%. Even in other experiments with the number training images more than test images, the image conditions have the same rotation but different scales, for classification accuracy had reached more than 93.68% with a total of 12 classes. Even if the drawing class studied is only 10 classes, the accuracy had reached more than 98%. The reliability of this algorithm in supporting batik image classification can continue to be improved in further research.