The Digit Recognition Using Local Projection Dependent Clustering

Abstract

Water companies utilize water meters to measure and calculate water usage bills. However, the current process employed by PDAM requires redundant resources, as it involves taking photos of each customer's house and having other officers read the numbers from the water meter images, resulting in inefficiency. The problem is further compounded by the neglect and improper maintenance of water meters, with some being buried in garbage or soil. Additionally, officers contribute to the challenges by capturing blurry and tilted photos, hindering the accurate reading of the water meter numbers. This study applies a water meter reading system by processing water meter photos and converting them into text using image processing methods to process images and Neural Networks to perform digit recognition. The image processing process includes steps such as (1) grayscale conversion, (2) gamma correction, (3) x-Histogram Projection, (4) White Temporal Ascent Accumulation, and (5) Peak Identification. Furthermore, image segmentation techniques are applied to enhance image quality and eliminate noise using clustering methods. The segmented images are then processed by a neural network to recognize the meter digits. The system achieves a digit recognition accuracy of 75.2%, despite encountering various technical and non-technical challenges during the water meter photo capture process.