The release of the free OpenCV 4.7 (Open Source Computer Vision Library) library has taken place, providing tools for processing and analyzing image content. OpenCV offers over 2500 algorithms, both classic and reflecting the latest advancements in computer vision and machine learning systems. The library's 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 for object recognition in photographs and videos (for example, face and human figure recognition, text, etc.), tracking the movement of objects and cameras, classifying actions in videos, image transformation, extracting 3D models, creating 3D spaces from images from stereo cameras, generating high-quality images through the combination of lower-quality images, searching for objects in images that resemble a given set of elements, applying machine learning methods, placing markers, identifying common elements across different images, and automating defect removal, such as red-eye effect.
Among the changes in the new release:
- Significant performance optimizations have been made to convolutions in the DNN (Deep Neural Network) module, implementing machine learning algorithms based on neural networks. The fast Winograd convolutions algorithm has been implemented. New ONNX (Open Neural Network Exchange) layers have been added: Scatter, ScatterND, Tile, ReduceL1, and ReduceMin. Support for the OpenVino 2022.1 framework and CANN backend has been added.
- The quality of QR code detection and decoding has been improved.
- Support for ArUco and AprilTag visual markers has been added.
- A neural network-based Nanotrack v2 tracker has been added.
- The Stackblur blurring algorithm has been implemented.
- Support for FFmpeg 5.x and CUDA 12.0 has been added.
- A new API for manipulating multi-page image formats has been proposed.
- Support for the libSPNG library for the PNG format has been added.
- Acceleraion using SIMD instructions has been implemented in libJPEG-Turbo.
- Support for H264/H265 has been implemented for the Android platform.
- All basic APIs for the Python language have been provided.
- A new universal backend for vector instructions has been added.
Source: opennet.ru
