Images in image libraries are sometimes oriented incorrectly. For example, a photograph taken with a camera held vertically may be displayed horizontally, or vice-versa. Further, users often capture images of documents; however, the resultant image can include distortions due to camera angle, poor lighting, etc. The captured image of a document often also includes objects outside the document boundary, e.g., a surface on which the document is placed. For some photos, automatic enhancements can enhance the quality of the image.
This disclosure applies machine learning techniques to detect if an image is that of a document, if the image is mis-rotated, if the image can benefit from automatic enhancement, etc. When such images are detected, enhancements such correction of rotation, cropping, distortion-removal, etc., are automatically suggested to the user, e.g., when the image is displayed. With user permission, an acceptance or dismissal of the suggestion is used as a training signal for the machine learning model. Enhancement suggestions are surfaced, e.g., as tappable or clickable buttons, when an image is being viewed and are applied upon user selection of the suggestion.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Pyszkiewicz, Karolina; Kazemi, Vahid; Xu, Hao; Song, Xuefeng; Cheng, Shinko; Cui, Jingyu; Dai, Shengyang; Mao, Kechun; Kulkarni, Mugdha; Sah, Shalini; Cannon, Marc; Liu, Yun; Timofeev, Aleksei; Gebhard, Megan; Wang, Nan; and Pillai, Jaishanker, "Image enhancement suggestions based on machine learning", Technical Disclosure Commons, (May 09, 2018)