The release of the free library OpenCV 4.7 (Open Source Computer Vision Library), which provides tools for processing and analyzing image content, has taken place. OpenCV provides more than 2500 algorithms, both classical and reflecting the latest advances in computer vision and machine learning systems. The library code is written in C++ and distributed under the BSD license. Bindings are prepared for various programming languages, including Python, MATLAB and Java.
The library can be used to recognize objects in photos and videos (for example, recognize faces and figures of people, text, etc.), track the movement of objects and the camera, classify actions on video, transform images, extract 3D models, form 3D space from images from stereo cameras, creating high-quality images by combining images of lower quality, searching for objects similar to the presented set of elements in the image, applying machine learning methods, placing markers, identifying common elements in different images, automatically eliminating defects such as red-eye .
Among the changes in the new release:
- A significant optimization of the performance of convolutions in the DNN (Deep Neural Network) module was carried out with the implementation of machine learning algorithms based on neural networks. Winograd's fast convolution algorithm has been implemented. Added new ONNX (Open Neural Network Exchange) layers: Scatter, ScatterND, Tile, ReduceL1 and ReduceMin. Added support for OpenVino 2022.1 framework and CANN backend.
- Improved quality of detection and decoding of QR codes.
- Added support for ArUco and AprilTag visual markers.
- Added tracker Nanotrack v2 based on neural networks.
- Implemented the Stackblur blur algorithm.
- Added support for FFmpeg 5.x and CUDA 12.0.
- A new API for manipulating multi-page image formats has been proposed.
- Added support for libSPNG library for PNG format.
- libJPEG-Turbo uses speedups using SIMD instructions.
- Support for H264/H265 has been implemented for the Android platform.
- All basic APIs for the Python language are provided.
- Added a new universal backend for vector instructions.
Source: opennet.ru