HP INCFollow


A document can be scanned via image scanners in two ways i.e., via flatbed or ADF. In flatbed, the user

cannot place a document in perfect orientation, as a result, the image acquired is slightly skewed.

Similarly, in the case of ADF, the skew in the image is introduced because of the mechanical nature of

how it pulls the page from its input tray. Hence, there is a need for an algorithm to straighten the skew

of the acquired image. This feature is very popular in scan solutions and almost always offered by

default because of the frequent occurrence of the problem. We propose a new method of skew

estimation and correction for images acquired via an image scanner. The method is not limited to just

documents but works for all scan use cases i.e., documents, photos, book pages, cards, etc. It can work

for use‐cases with very little to no text if there is a representative pattern w.r.t orientation of content of

an image. It is in‐line in nature i.e., it only needs a band of a foreground image to estimate the skew of

the image. The solution is scalable w.r.t resources of the user i.e. compute and memory as you can

adjust its output precision of skew estimation based on resources at hand.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.