HDR image synthesis compose consists of the following steps:
Step0 Image Registration, accurate registration of the input image is required in the event of an image motion
Step1 Recover different exposure images (E1, E2, E3 ...) Radiometric images (S1, S2, S3 ...)
Step2 calculation S1, S2, S3. The weight value W1 (x, y), W2 (x, y), W3 (x, y), to synthesize the last picture
Whether it's a multi-frame synthetic HDR or a sensor with a special exposure pattern, follow the steps above. Expand the Introduction:
Step0 Image Registration, referring to the realization of optical flow, this step is mainly used in multi-frame different exposure image synthesis situation, is the basis of the next two steps
Step1 Restores the irradiance image, which is the original irradiance of the pixel based on the exposure time (in most cases, the high dynamic range is obtained by exposure time) and the CMOS grayscale value.
The corresponding relationship between irradiance and exposure time and CMOS gray level.
E (i,j) = Func (T (i) x i (J))
E (I,J) is the first exposure, the exposure time is T (i), the actual irradiance of the first J pixels of the image I.
The parameters of this curve can be obtained by combining the least squares according to the calibration method.
Theoretically, if there is no noise, then according to the STEP1, it can be fused into an HDR image (with other images to compensate for an over-exposure in an image and the lack of exposure, so that the calculated irradiance is inaccurate points). Of course, the reality is not so, considering the calculated irradiance is a certain degree of uncertainty, we need the least uncertainty. So we need step2.
Step2 calculates a radiometric image to synthesize weights at each pixel point of an image, generating n-weighted graphs.
There are various ways to generate weights, mainly considering the gradient of the point neighborhood, brightness, motion or not (short and long exposure), to investigate the degree of the radiation generated by the image reliability, and then determine the weight map.
Next time will speak tone mapping tone mapping, this is mainly to get the HDR image, how to display on the display, after all, HDR image bit is higher, and the display range is limited. High dynamic range compression to ground dynamic range. Includes global mappings and local mappings.
HDR-related topic has a lot of knowledge, including the evaluation of HDR algorithms.
HDR Algorithm (i)