# US 2018/0322357 A1 · System and method for image re-orientation > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2018/0322357 A1 - Document ID: US20180322357A1 - Status: Published application - Kind: A1 - Application number: 15/990,570 - Filed: 2018-05-25 - Published: 2018-11-08 - Earliest claimed: 2017-05-05 - Assignees: Beijing Didi Infinity Technology and Development Co., Ltd. - Applicants: Beijing Didi Infinity Technology and Development Co., Ltd. - Inventors: Xin CHEN; Min LI - Web page: https://patentgrep.com/patent/US20180322357A1 ## CPC - G06T 7/73: using feature-based methods - G06F 18/2148: {characterised by the process organisation or structure, e.g. boosting cascade} - G06T 3/60: Rotation of whole images or parts thereof - G06V 10/242: {by image rotation, e.g. by 90 degrees} - G06V 30/418: Document matching, e.g. of document images - G06T 2207/10004: Still image; Photographic image - G06T 2207/20081: Training; Learning - G06T 2207/30176: Document - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING ## Abstract Systems and methods are provided for image re-orientation. A computing system for image re-orientation may comprise one or more processors and a memory for storing instruction that, when executed by the processors, cause the system to perform: obtaining positive samples including text of a predetermined type in a predetermined orientation range and negative samples not including text of the predetermined type in the predetermined orientation range; training a model with the obtained positive and negative samples; receiving an image in an original orientation; applying the trained model to the received image; in response to determining that the received image does not comprise any text of the predetermined type in the predetermined orientation range, rotating the image by a predetermined angle and re-applying the trained model to the rotated image to determine if the rotated image comprises at least some text of the predetermined type in the predetermined orientation range. ## Claims ### Claim 1 A computing system for image re-orientation, comprising: a memory storing instructions; and one or more processors in communication with the memory, wherein when executing the instructions, the one or more processors are configured to cause the system to: obtain positive and negative samples, wherein a positive sample includes text of a predetermined type in a predetermined orientation range, and a negative sample does not include text of the predetermined type in the predetermined orientation range; train a model with the obtained positive and negative samples; receive an image in an original orientation; apply the trained model to the received image to determine if the received image comprises at least some text of the predetermined type in the predetermined orientation range consistent with the positive samples; in response to determining that the received image does not comprise any text of the predetermined type in the predetermined orientation range, rotate the image by a predetermined angle and re-apply the trained model to the rotated image to determine if the rotated image comprises at least some text of the predetermined type in the predetermined orientation range, until determining that the rotated image comprises at least some text of the predetermined type in the predetermined orientation range consistent with the positive samples or that the received image is rotated a predetermined number of degrees; and in response to determining that the received or rotated image comprises at least some text of the predetermined type in the predetermined orientation range, retain the image's current orientation. ### Claim 2 (depends on 1) The system of claim 1, wherein the image at least includes document information. ### Claim 3 (depends on 2) The system of claim 2, wherein the document is a driver's license. ### Claim 4 (depends on 1) The system of claim 1, wherein: the predetermined type comprises at least one of a title, a serial number, or a date. ### Claim 5 (depends on 1) The system of claim 1, wherein: the predetermined orientation range comprises a configurable degree of deviation from an upright orientation. ### Claim 6 (depends on 5) The system of claim 5, wherein: the predetermined orientation range comprises 15 degrees from an upright orientation. ### Claim 7 (depends on 1) The system of claim 1, wherein: the positive samples include image portions comprising at least one of title, serial number, or date oriented within 15 degrees from an upright orientation; and the negative samples include image portions that do not have title, serial number, or date oriented within 15 degrees from the upright orientation. ### Claim 8 (depends on 1) The system of claim 1, wherein to rotate the received image by the predetermined angle and to re-apply the trained model to the rotated image, until determining that the rotated image comprises at least some text of the predetermined type in the predetermined orientation range consistent with the positive samples or that the received image is rotated a predetermined number of degrees, the one or more processors are further configured to cause the system to perform—the following one or more rotations in a same direction: rotating the image by 90 degrees from the original orientation and re-applying the trained model to determine if the rotated image comprises at least some text in the predetermined orientation range; rotating the image by 180 degrees from the original orientation and re-applying the trained model to determine if the rotated image comprises at least some text in the predetermined orientation range; and rotating the image by 270 degrees from the original orientation and re-applying the trained model to determine if the rotated image comprises at least some text in the predetermined orientation range. ### Claim 9 An image re-orientation method, comprising: obtaining positive and negative samples, wherein a positive sample includes text of a predetermined type in a predetermined orientation range, and a negative sample does not include text of the predetermined type in the predetermined orientation range; training a model with the obtained positive and negative samples; receiving an image in an original orientation; applying the trained model to the received image to determine if the received image comprises at least some text of the predetermined type in the predetermined orientation range consistent with the positive samples; in response to determining that the received image does not comprise any text of the predetermined type in the predetermined orientation range consistent with the negative samples, rotating the image by a predetermined angle and re-applying the trained model to the rotated image to determine if the rotated image comprises at least some text of the predetermined type in the predetermined orientation range, until determining that the rotated image comprise at least some text of the predetermined type in the predetermined orientation range consistent with the positive samples or that the received image is rotated a predetermined number of degrees; and in response to determining that the received or rotated image comprises at least some text of the predetermined type in the predetermined orientation range, retaining the image's current orientation. ### Claim 10 (depends on 9) The method of claim 9, wherein the image at least includes document information. ### Claim 11 (depends on 10) The method of claim 10, wherein the document is a driver's license. ### Claim 12 (depends on 9) The method of claim 9, wherein: the predetermined type comprises at least one of a title, a serial number, or a date. ### Claim 13 (depends on 9) The method of claim 9, wherein: the predetermined orientation range comprises a configurable degree of deviation from an upright orientation. ### Claim 14 (depends on 13) The method of claim 13, wherein: the predetermined orientation range comprises 15 degrees from an upright orientation. ### Claim 15 (depends on 9) The method of claim 9, wherein: the positive samples include image portions comprising at least one of title, serial number, or date oriented within 15 degrees from an upright orientation; and the negative samples include image portions that do not have title, serial number, or date oriented within 15 degrees from the upright orientation. ### Claim 16 (depends on 9) The method of claim 9, wherein rotating the received image by the predetermined angle and re-applying the trained model to the rotated image, until determining that the rotated image comprises at least some text of the predetermined type in the predetermined orientation range consistent with the positive samples or the received image is rotated a predetermined number of degrees comprises performing the following one or more rotations in a same direction: rotating the image by 90 degrees from the original orientation and re-applying the trained model to determine if the rotated image comprises at least some text in the predetermined orientation range; rotating the image by 180 degrees from the original orientation and re-applying the trained model to determine if the rotated image comprises at least some text in the predetermined orientation range; and rotating the image by 270 degrees from the original orientation and re-applying the trained model to determine if the rotated image comprises at least some text in the predetermined orientation range. ### Claim 17 An image re-orientation method, comprising: obtaining positive and negative samples, wherein a positive sample includes text of a predetermined type in a predetermined orientation range, and a negative sample does not include text of the predetermined type in the predetermined orientation range; training a model with the obtained positive and negative samples; receiving an image in an original orientation; applying the trained model to the received image to determine if the received image is consistent with the positive samples or negative samples; in response to determining that the received image is consistent with the negative samples, rotating the image by a predetermined angle and re-applying the trained model to the rotated image to determine if the rotated image is consistent with the positive or negative samples or if the image is rotated a predetermined a predetermined number of degrees; and in response to determining that the received or rotated image is consistent with the positive samples, retaining the image's current orientation. ### Claim 18 (depends on 17) The method of claim 17, wherein rotating the received image by the predetermined angle and re-applying the trained model to the rotated image comprises performing the following one or more rotations in a same direction: rotating the image by 90 degrees from the original orientation and re-applying the trained model to determine if the rotated image is consistent with the positive or negative samples; rotating the image by 180 degrees from the original orientation and re-applying the trained model to determine if the rotated image is consistent with the positive or negative samples; and rotating the image by 270 degrees from the original orientation and re-applying the trained model to determine if the rotated image is consistent with the positive or negative samples. ### Claim 19 (depends on 17) The method of claim 17, wherein: the positive samples include image portions comprising at least one or title, serial number, or date orientated within 15 degrees from an upright orientation; and the negative samples include image portions that do not have title, serial number, or date orientated within 15 degrees from the upright orientation. ### Claim 20 (depends on 19) The method of claim 19, wherein: the positive samples include driver's licenses.