Conspect for OpenCV: Introduction

03 Sep, 2025

Some notable notes from https://docs.opencv.org/4.x/d1/dfb/intro.html.

1. Automatic memory management
   • ref counting
   • cv::Mat and others contain ref to buffer
     • cv::Mat doc
   • cv::Mat::clone to copy buffer
   • use cv::Ptr for high-level class to include OpenCV ref-counting
     • cv::Ptr doc (for 3.4 though)
   • if cv::Mat instance is reused with different dimensions (e.g. in cap >> frame), storage is automatically reallocated
2. Saturation arithmetics
   • use cv::saturate_cast<cv::uchar> to cast a value into the acceptable range
3. Fixed pixel types
   • because OpenCV limits the use of templates use CV_ types for element
   • CV_8U and so on are for scalars
   • CV_8UC2 or CV_8UYC(2) or CV_MAKETYPE(img.depth(), 1) is for 2-channel values
   • more complex elements are not possible in OpenCV arrays
   • algorithms are limited in what array element types they handle
4. InputArray and OutputArray
   • use cv::Matx<...> - for small matrixes
   • InputArray and OutputArray are any supported containers Mat, Matx, Vec, Scalar, or std::vector
   • use cv::noArray() as a dummy for optional parameters
5. Error handling
   • in case of error, all intermediate buffers are automatically deallocated
6. OpenCV functions or methods of different instances can be used from different threads
   • you cannot call a method of the same instance from two different threads simultaneously, as the internal data is not protected (unless you synchronize access, of course)

Maybe explore next:

• image pixel encoding - 8- or 16-bit per channel
• why saturate_cast is ok - some information is lost?

#conspect #opencv