Automated Line Segmentation for Gurmukhi Text Recognition System

In a Text Recognition system, segmentation is a critical and difficult phase. When words in the target language cover more than one zone, such as in Gurmukhi, it becomes more challenging. Moreover, automated systems are becoming more popular in this modern age due to their ability to save time and labour. This research also looks at the automation process for text recognition systems, in which lines are segmented automatically without the need for external assistance, particularly in the Gurmukhi language. Earlier, only a few solutions were developed, but they were not evaluated on huge datasets, which does not reflect the efficiency of the existing systems. So, in order to produce efficient results, an automated line segmentation strategy ‘AutoLineSeg’ is proposed in this work, which uses the projection profile method along with some pre-processing. To offer reliable data for segmentation, this proposed system additionally does skew correction. The testing of the system is done on the dataset with 1000 images of typewritten Gurmukhi samples. The proposed method's accuracy is calculated, and the findings reveal that the system obtained around 91% accuracy, which is sufficient for a text recognition system.