Pengaruh Jumlah Layer Simetris Terhadap Akurasi Sistem Handwriting Recognition Offline

Abstrak

Image recognition field of science has developed, among others, in handwriting recognition or handwriting recognition, both online and offline. Several previous studies mentioned that the accuracy of handwriting recognition system can be improved by using a combination of the number of input layer, hidden layer and output layer appropriately. But it is not yet in a concrete formulation. This is due to the structure of Neural Network used are also very varied, although equally composed of input layer, hidden layer and ourput layer. In this study will be sought influence the amount of the layer that is symmetrical to the level of accuracy. Selection of the amount of the symmetrical layers based on that all structures Neural Network consists of input layer, hidden layer and ourput layer. Tests performed on the input data 5 that is a handwriting from 5 different audience by using a symmetrical 10 the number of layers between 121 to 400. As well pembading done tests on the same data as the number of input layer, hidden layer and output layer varying between 196 So the total to 400. The test is as much as 100 times. In this study, the total amount of time the process is not a consideration. The amount of the symmetric layer produces a better accuracy rate than the use of the number of layers that are not symmetrical, although not very significant. Tertiary difference value obtained accuracy is 8%.