No-reference image quality assessment for iterative batch video analysis
Introduced here are technologies for examining content generated by electronic devices in real time to optimize the quality of the content. The content may be examined in batches to address some of the drawbacks of real-time analysis. For instance, a series of videos may be examined to collect data on how well security system(s) that are presently employed are working. Each security system can include one or more electronic devices, such as cameras or microphones, and parameters of the electronic devices can be altered to improve the quality of content generated by the electronic devices.
1. A computer-implemented method for operating a camera, comprising:
receiving video captured by the camera having a parameter set to a first value based on pre-launch training of a reference-less model;
establishing a quality metric for the video using the reference-less model by determining a difference in statistical properties of the video caused by distortions captured in the video,
wherein the reference-less model is trained on a database of videos with known distortions,
wherein the quality metric comprises a local image structure and a luminance for the video, and
wherein the reference-less model comprises scene statistics to quantify a loss of quality in the videos from the distortions;
receiving input indicative of an instruction to alter a value of the parameter; and
transmitting the instruction to the camera to cause the parameter to be set to a second value.
2. The method of claim 1 , wherein the second value is based on the quality metric for the video captured by the camera having the parameter set to the first value.
3. The method of claim 1 , comprising comparing the video to an optimized reference video of an environment previously observed by the camera for establishing the quality metric.
4. The method of claim 3 , comprising determining the second value based on a past value of the parameter used by the camera to capture the optimized reference video.
5. The method of claim 1 , wherein the reference-less model considers as input pixel data or parametric data of each video in the database of videos.
6. The method of claim 1 , wherein the first and second values are based on at least one of:
a constraint specified in a profile; or
a metric produced by an image quality assessment operation.
7. The method of claim 1 , wherein the quality metric is a numeric representation of hue, blurriness, contrast, saturation, brightness, or any combination thereof.
8. An electronic device for operating a camera, comprising:
a network interface for communicating with the camera over a wireless communication channel;
a processor; and
a memory having instructions stored thereon that, when executed by the processor, cause the processor to:
receive video captured by the camera having a parameter set to a first value based on pre-launch training of a reference-less model;
establish a quality metric for the video using the reference-less model by determining a difference in statistical properties of the video caused by distortions captured in the video,
wherein the reference-less model is trained on a database of videos with known distortions,
wherein the quality metric comprises a local image structure and a luminance for the video, and
wherein the reference-less model comprises scene statistics to quantify a loss of quality in the videos from the distortions;
receive input indicative of an instruction to alter a value of the parameter; and
transmit the instruction to the camera to cause the parameter to be set to a second value.
9. The electronic device of claim 8 , wherein the second value is based on the quality metric for the video captured by the camera having the parameter set to the first value.
10. The electronic device of claim 8 , wherein the instructions cause the processor to compare the video to an optimized reference video of an environment previously observed by the camera for establishing the quality metric.
11. The electronic device of claim 10 , wherein the instructions cause the processor to determine the second value based on a past value of the parameter used by the camera to capture the optimized reference video.
12. The electronic device of claim 8 , wherein the reference-less model considers as input pixel data or parametric data of each video in the database of videos.
13. The electronic device of claim 8 , wherein the first and second values are based on at least one of:
a constraint specified in a profile; or
a metric produced by an image quality assessment operation.
14. The electronic device of claim 8 , wherein the quality metric is a numeric representation of hue, blurriness, contrast, saturation, brightness, or any combination thereof.
15. A computer system for operating a camera, comprising:
a processor; and
a memory having instructions stored thereon that, when executed by the processor, cause the processor to:
receive video captured by the camera having a parameter set to a first value based on pre-launch training of a reference-less model;
establish a quality metric for the video using the model by determining a difference in statistical properties of the video caused by distortions captured in the video,
wherein the model is trained on a database of videos with known distortions,
wherein the quality metric comprises a local image structure and a luminance for the video, and
wherein the model comprises scene statistics to quantify a loss of quality in the videos from the distortions;
receive input indicative of an instruction to alter a value of the parameter; and
transmit the instruction to the camera to cause the parameter to be set to a second value.
16. The computer system of claim 15 , wherein the second value is based on the quality metric for the video captured by the camera having the parameter set to the first value.
17. The computer system of claim 15 , wherein the instructions cause the processor to compare the video to an optimized reference video of an environment previously observed by the camera for establishing the quality metric.
18. The computer system of claim 15 , wherein the model is a Blind/Reference-less Image Spatial Quality Evaluator (BRISQUE) model or a Natural Image Quality Evaluator (NIQE) model.
19. The computer system of claim 15 , wherein the pre-launch training uses a Mean-Squared Error (MSE) method.