Oscar Software Framework Manual Calibration Module
Flash partition /calib to read calibration data.
The calibration module provides functionality to correct image sensor data from non-uniformities. No correction is applied for host compilation.
Image sensors suffer from multiple noise/mismatch impacts. A common approach is to calibrate each individual sensor in terms of fixed-pattern-nosise (FPN), photo response non-uniformity (FPNU) and hotpixel effects. The accquired parameter set is stored persistent memory.
The calibration module supports the mentioned image correction methods. The calibration data is loaded from a dedicated flash partition /calib.
Host: No correction is applied for host compilation.
The slope correction assumes linear relation between illumination and pixel response. Two non-uniformities are observed. The blacklevel response of all pixels has a gaussian distribution. Once this property is corrected also the greylevel (3/4 of max illumination) response is distributen.
The image from the sensor is processed according to this equation: img' = (img - offset) · gain
With offset is FPN correction data and gain is PRNU factors. In order to get the a undistorted gaussian response for the blacklevel the sensor must be configured.
Hot pixel correction is applied after the slop correction. Affected pixels are replaced by the linear interpolation result of the four neighbour pixels, as shown below.
The calibration file has followings sections:
- magic word, used as identification for the flash update application (imgupd)
- fpn data, 5bit per Pixel
- prnu data, 11bit per Pixel
- hotpixel: number of hotpixels, coordinate pair for each hotpixel
The 5bit FPN allows to compensate for fixed pattern in the range of [0, 32). For optimal calibration results the gaussian response of the sensor is inside this area.
The 11 Bit PRNU data is interpreted as gain factor in fixed point format xxx.yyyyyyyy. This allows to encode gain correction: [0, 16). This wide range of gain corrections is usefull for calibration of normal scenes instead of a homogenous grey image.
Hotpixel coordinates must be sorted so that pixels on lines at the top come before pixels at the bottom an that pixels at the left of a line come before pixels at the right of the same line.
Calibration File Specification:
Encoding of FPN and PRNU values:
Following code segment demonstrates the usage of the bmp module. For the sake of simplicity, error checking is neglected.
OscClbSetupCalibrate(OSC_CLB_CALIBRATE_FPN_PRNU, FALSE); /* (2) */ OscCamReadPicture(..) /* (3) */ /* optional */ OscCamSetupCorrectionCallback( myOwnCorrectionFxn) /* (4) */
- Load the module. Do not load module, in case no image correction is required. This save boot time and memory resources.
- Setup the prefered image correction method.
- slope correction: offset only (OSC_CLB_CALIBRATE_FPN), gain only (OSC_CLB_CALIBRATE_PRNU), offset and gain (OSC_CLB_CALIBRATE_FPN_PRNU)
- hotpixel correction
- Read a next picture from the camera. Image correction is automatically applied as configured.
- Optionally a own correction function can to supplied. It is called instead of the build in functions and configured methods.