SDR to HDR conversion method

Abstract

The invention discloses an SDR to HDR conversion method, relates to an inverse tone mapping model based on machine learning, and belongs to the field of image processing. The SDR to HDR conversion method disclosed in the invention comprises the steps: predicting a high-brightness threshold value of a single-lens image through a machine learning method, and dividing the image into a diffuse reflection region and a high-light reflection region through the high-brightness threshold value; enabling the brightness of the dark areas of different lenses to be relatively stable, so that the brightness threshold of the dark areas is given according to experience; respectively establishing different brightness mapping functions for the dark part region, the diffuse reflection region, the high reflection region and the transition region among different brightness regions; and in a small interval which is close to the dark part threshold value and takes the high brightness threshold value as the center, taking a cubic spline interpolation function as a brightness mapping function, so that the brightness among the brightness areas is in smooth transition. The method can be applied to the fields of video popularization, televisions and movies and the like.

Description

technical field [0001] The invention relates to an inverse tone mapping model based on machine learning, and belongs to the field of image processing. Background technique [0002] In recent years, high-dynamic range (High-Dynamic Range, HDR for short) display technology has gradually emerged in the fields of film and television, games, and virtual reality. HDR display devices have higher peak brightness and contrast ratios than conventional displays, allowing HDR images to be displayed naturally. For reasons such as technology and production cost, the current market is still dominated by Standard Dynamic Range (SDR) images. According to general SDR content production specifications, the maximum brightness displayed by SDR images is 100nit, while the peak brightness of most home-level and production-end HDR displays on the market is generally around 1000nit. Simply increasing the brightness of the SDR image to the brightness of the HDR image will cause abnormal image contr...

AI Technical Summary

Problems solved by technology

Although the model-driven method can intuitively describe the law through curves or mathematical formulas, since a function model may not be suitable for all types of images, the disadvantage of this type of method is that the scope of application is relatively narrow, and the need to Manually adjust the parameters in the model multiple times, otherwise it is difficult to obtain high-quality images

Compared with model-driven methods, this type of method has a wider range of applications. However, this type of method needs to be trained and fitted on a large amount of data, a large number of training samples need to be collected, and a large amount of training time is required. In addition, Some unnecessary artifacts may be generated during image processing, affecting image quality

Images