Study comparison of color channel to median filter in wayang image using convolutional neural network algorithm

Industry 4.0 has a tremendous impact and influence on the development of information technology. Various applications of information technology that were previously carried out conventionally are now undergoing drastic changes. One area currently being developed is the introduction of an object automatically by the system or often called computer vision. The rapid development of computer vision has resulted in many algorithms being developed and modified to find optimal performance. One of the widely used algorithms is CNN (Convolutional Neural Network). From several previous studies related to comparing several algorithms, CNN is dominant and superior to other algorithms. The accuracy of object recognition is influenced by the algorithm and many factors that influence it. One of them is the type of image used and the treatment of the image before analysis. In general, the widely used image in computer vision is an RGB image consisting of 3 color channels. Each channel has a characteristic pixel value and different degrees of gray. There are several types of data treatment before analysis. One of them is the filtering process which aims to improve the quality of an image using a median filter. Therefore, the authors are interested in comparing three color channels to the preprocessed treatment of wayang images using the CNN algorithm to obtain the best scenario combination in terms of the accuracy obtained. From several scenarios, the best scenario is the green channel without the use of filtering with an accuracy of 96.25%. Because based on previous research literature, the green channel contains the slightest noise, while the filtering method aims to reduce the noise in an image. Thus, median filtering is less effective for green channels. Increasing the value of the median filter parameter to reduce noise will have a side effect of blurring the image and also disrupting the object recognition process.