This is a web-based demonstration of the OCRMPD system. Upload an image of Myanmar text to see a simulation of the recognition process based on the proposed algorithms.
The system first isolates individual characters. A vertical projection histogram (X-Y cut) makes initial cuts. Then, a crucial Structural Analysis step refines these cuts, checking pixel density and connected components (CCs) to accurately segment complex, overlapping Myanmar characters. Red boxes show the detected segment boundaries on the image below.
Segments from the original image will be displayed here after processing.
After segmentation, each character image is normalized. A hybrid statistical method extracts unique features using Zone Density (pixel density in 3 horizontal zones) and Projection Area (area from top/bottom/left/right profiles). This creates a feature vector for each character.
Finally, a Hierarchical Multi-class Support Vector Machine (SVM) classifier receives the feature vector. The hierarchical model efficiently navigates the large and visually similar Myanmar character set to determine the final text output.