sol3::isp Namespace Reference

Classes
struct	AutoWhiteBalanceConfig

Functions
std::optional< msg::WhiteBalanceGains >	estimateWhiteBalance (cv::Mat const &bgr_image, cv::Rect const &roi, AutoWhiteBalanceConfig const &config={})
void	yuvToBgr (void const *data, size_t size, core::msg::BufferInfo const &info, cv::Mat &bgr)
void	yuvToBgr (core::IBufferConst const &buffer, cv::Mat &bgr)

Function Documentation

estimateWhiteBalance()

std::optional<msg::WhiteBalanceGains> sol3::isp::estimateWhiteBalance	(	cv::Mat const &	bgr_image,
		cv::Rect const &	roi,
		AutoWhiteBalanceConfig const &	config = {}
	)

Estimate white balance gains from a BGR image, scoped to roi (clipped to the image bounds).

The estimator assumes the gray-world hypothesis: averaged over a "natural" scene under truly white illumination, the mean of every channel is equal. Any inequality is treated as an illuminant color cast (or sensor-response mismatch) and corrected by per-channel scaling. Computed as: r_gain = mean_g / mean_r g_gain = 1.0 b_gain = mean_g / mean_b

Green is pinned at 1 because:

• White balance only constrains channel ratios, so the absolute scale is a free parameter that must be normalized somehow.
• Bayer sensors have twice the green photosites of R/B, making G the lowest-noise channel; using it as the reference avoids amplifying noise.
• BT.601 luminance weights green at ~0.587, so an unmodified G approximately preserves apparent brightness through the WB pass.
• It is the de-facto convention for camera ISPs.

Limitations: the gray-world hypothesis fails on scenes dominated by a single hue (e.g. a green field), which would otherwise produce huge gains on the under-represented channels. The two config knobs guard against this:

• config.min_channel_mean rejects ROIs where any channel mean drops below the threshold.
• config.max_gain clamps R and B to a finite ceiling.

The returned gains satisfy G == 1.0 and 0 <= R, B <= config.max_gain. Returns std::nullopt for an empty image, an empty/out-of-bounds ROI, or insufficient signal in any channel.

yuvToBgr() [1/2]

void sol3::isp::yuvToBgr	(	core::IBufferConst const &	buffer,
		cv::Mat &	bgr
	)

Convenience overload that pulls data, size, and info from the buffer.

yuvToBgr() [2/2]

void sol3::isp::yuvToBgr	(	void const *	data,
		size_t	size,
		core::msg::BufferInfo const &	info,
		cv::Mat &	bgr
	)

Decode a YUV buffer into bgr, sized to the active image extent (not the padded stride). Format and dimensions come from info; only BufferFormat_YUV_NV12 is currently supported.

bgr is reused across calls. Pass a persistent cv::Mat to avoid the per-call allocation that cv::cvtColor's allocator would otherwise do.

Preconditions (asserted via SOL3_EXPECT):

• info.format() is a supported YUV format
• width and height are non-zero, height is even
• stride_w >= width, stride_h >= height
• size is large enough for the layout