- Number
- 11494584
- Published
- 2022-11-08
- Filed
- 2021-01-12
- Assignee
- Disney Enterprises, Inc.
- Inventors
- Doggett; Erika Varis, Nguyen; David T., Qi; Erick Keyu, Zhong; Yingying, Cui; Weichu, Wolak; Anna M.
- CPC
- G06F18/2148; G06F18/41; G06T7/0002; G06T7/0004; G06V10/25; G06V10/82; G06V40/20
- Verdict
- Set aside generic image QC/anomaly detection
- Source
- Google Patents · FreePatentsOnline
Abstract
A system includes a hardware processor and a memory storing a software code including a predictive model. The hardware processor executes the software code to receive an input including an image having a pixel anomaly, and image data identifying the location of the pixel anomaly in the image. The software code uses the predictive model to extract a global feature map of a global image region of the image, the pixel anomaly being located within the global image region; to extract a local feature map of a local image region of the image, the pixel anomaly being located within the local image region and the local image region being smaller than the global image region; and to predict, based on the global feature map and the local feature map, a distraction level of the pixel anomaly within the image.
Background
BACKGROUND (1) Pixel errors in images occur with regularity but can be difficult and costly to correct. For example, pixel anomalies in video frames can be introduced by many different processes within a video production pipeline. A final quality procedure for correcting such errors is typically done before the video undergoes final release, and in the conventional art that process is usually performed by human inspectors. Due to its reliance on human participation, pixel error correction is expensive and time consuming. (2) However, not all pixel errors require correction. For example, depending on its position within an image, its visual impact relative to other features in its local environment within the image, as well as the visual qualities of the image as a whole, some pixel anomalies may be highly distracting, while others may be less so, and still others may reasonably be disregarded without significantly affecting an intended esthetic of the image. That is to say, not all pixel errors are of equal importance. Accordingly, there is a need in the art for an automated approach to predicting the noticeability of pixel errors in an image. SUMMARY (3) There are provided systems and methods for performing automated prediction of pixel error noticeability, substantially as shown in and described in connection with at least one of the figures, and as set forth more completely in the claims.
Claims
1. A system comprising: a hardware processor; and a system memory storing a software code including a predictive model; the hardware processor configured to execute the software code to: receive an input, the input including an image having a pixel anomaly, and an image data identifying a location of the pixel anomaly in the image; extract, using the predictive model, a global feature map of a global image region of the image, the pixel anomaly being located within the global image region; extract, using the predictive model, a local feature map of a local image region of the image, the pixel anomaly being located within the local image region, the local image region being smaller than the global image region; and predict, using the predictive model and based on the global feature map and the local feature map, a distraction level of the pixel anomaly within the image. ||
10. A method for use by a system having a hardware processor and a system memory storing a software code including a predictive model, the method comprising: receiving an input, by the software code executed by the hardware processor, the input including an image having a pixel anomaly, and an image data identifying a location of the pixel anomaly in the image; extracting, by the software code executed by the hardware processor and using the predictive model, a global feature map of a global image region of the image, the pixel anomaly being located within the global image region; extracting, by the software code executed by the hardware processor and using the predictive model, a local feature map of a local image region of the image, the pixel anomaly being located within the local image region, the local image region being smaller than the global image region; and predicting, by the software code executed by the hardware processor and using the predictive model, based on the global feature map and the local feature map, a distraction level of the pixel anomaly within the image. ||
18. A method for use by a training platform including a hardware processor, a memory storing a training software code, and a display to generate a just-noticeable difference (JND) based training dataset for use in training a predictive model, the method comprising: for each one of a plurality of images that includes a pixel error: displaying to a human observer for a period of time, by the training software code executed by the hardware processor and using the display, the one of the plurality of images with the pixel error at a first pixel size; determining, by the training software code executed by the hardware processor, whether the human observer notices the pixel error at the first pixel size; assigning a first distraction level to the pixel error, by the training software code executed by the hardware processor, when determined that the human observer notices the pixel error at the first pixel size; displaying to the human observer for the period of time, by the training software code executed by the hardware processor and using the display, the one of the plurality of images with the pixel error at a second pixel size larger than the first pixel size, when determined that the human observer does not notice the pixel error at the first pixel size; determining, by the training software code executed by the hardware processor, when determined that the human observer does not notice the pixel error at the first pixel size, whether the human observer notices the pixel error at the second pixel size; and assigning a second distraction level lower than the first distraction level, to the pixel error, by the training software code executed by the hardware processor, when determined that the human observer notices the pixel error at the second pixel size.