Comparison of Edge Detection Method in Case of Blood Pattern Recognition Using Backpropagation Algorithm

Abstract

There are 4 types of blood: A, B, O, and AB. So far, the process of checking blood type depends on the officer’s work accuracy. To keep the validity of the results, a system is needed to help humans to recognize the blood types. This recognition can be done by computers by applying the method of blood pattern recognition through an image. The data domain of this study is a scan of blood type checks obtained from PMI Yogyakarta City. A total of 54 images were used in the training and recognition process. The image used in .bmp extension with a size of 400 x 200 pixels. Before the recognition process, first execute the preprocessing image, that is convert the image to grayscale image. The next process is edge detection with a Sobel operator or Prewitt operator. The use of these two operators aim to determine the optimal operator for recognition of blood type case. After the edge detection process, the image is converted to binary so it can be processed by feature extraction. The last step is the implementation of artificial neural network backpropagation algorithm with bipolar sigmoid activation function for hidden layer and linear activation for output. As a result, the optimal neural network architecture is three hidden layers with each hidden layer having three nodes. The optimal value for the mean squared error parameter is 1e-1 or 0.1, epoch 1000 and learning rate 0.01. In this study, Sobel operator was better than Prewitt operator in introducing blood type types. When viewed from the difference in processing time, the Prewitt operator is slightly faster than the Sobel operator with a difference of 0.000052 seconds. From 39 training data and 14 test data, the percentage of success in the recognition of blood type was 92.86%.