Adaptive document image thresholding using foreground and background clustering

Two algorithms for document image thresholding are presented, that are suitable for scanning document images at high-speed. They are designed to operate on a portion of the image while scanning the document, thus, they fit a pipeline architecture and lend themselves to real-time implementation. The first algorithm is based on adaptive thresholding and uses local edge information to switch between global thresholding and adaptive local thresholding determined from the statistics of a local image window. The second thresholding algorithm is based on tracking the foreground and background levels using clustering based on a variant of the K-means algorithm. The two approaches may be used independently or may be combined for improved performance. Results are presented illustrating the algorithms' performance for document and pictorial images.