Text-Guided Real-World Image Restoration

Capturing images in limited conditions (such as low-light, high dynamic range, etc.) is hard and the resulting images are not good. Since the camera is limited in such scenarios, capturing such scenes becomes difficult.
Using a novel image processing method with artificial intelligence (AI) technology the user can talk to the camera and describe the captured scene to improve and restore the images even under hard imaging conditions.

The most popular cameras these days are smartphone cameras which have become much smaller to reduce smartphone size, while imaging performance is compromised. To overcome this poor imaging system, massive algorithmic effort is made by software to improve the captured images. In strongly degraded images there is a loss of information that can’t be retrieved and as such the reconstructed results are poor. Smartphone cameras are only one use case and the technology can be integrated into any camera system, since cameras do not operate properly under such limited conditions.

We present an image processing algorithm conditioned by an input text from the photographer. The algorithm processes the captured image to improve the image using text input. The input text helps the algorithm in hard or limited imaging conditions to recover better images for the user. The input text is essential for the algorithm to recover and enhance the image. Without the input text when the captured image is too much degraded, the algorithm can’t recover a better image from it.

Since currency existing algorithms do not use text input their capabilities are limited since they do not know what is the content of the captured scene and what should they reconstruct. Using our novel method, the additional information given to the algorithm in the form of text or speech enables the algorithm to perform better for the specific domain or objects in the captured scene. Such interaction with the user while capturing or improving photos is unique and novel and can be the new standard for image capturing using smartphones in the future because of the promising results.

Provisional patent application

Sign up for
our events

    Life Science