| Summary: | Optical Character Recognition (OCR) is an application in pattern recognition to recognize characters on the digital image. In this study, OCR is used to recognize serial number on medical gas cylinders. Medical gas cylinders have serial numbers written with paint on the body of the gas cylinder. Therefore, the serial numbers is susceptible to noise such as paint cracks on serial numbers and background. In addition, there are serial numbers written with non-standard mold so the shapes of its character like a handwriting characters.
The method used in the system are image enhancement, character segmentation and serial number recognition. Image enhancement is done by applying bilateral filter to refine image and sharpen image edges. Character segmentation is done by thresholding serial numbers and background color labels obtained from K-means clustering. Serial number recognition is done by applying back propagation neural network on characters serial number obtained from character segmentation.
The tests conducted with 20 serial number of medical gas cylinders image samples. The test results showed 95,05% detection accuracy with 1,98% error and 91,09% recognition accuracy. Accuracy mainly influenced by noise such as plate conditions, false positives, and completeness of the background.
|
|---|