In this article, we will organize the Python libraries commonly used in computer vision projects. If you want to enter the field of computer vision, you can first understand the libraries introduced in this article, which will be very helpful for your work.
1、PIL/Pillow
Pillow is a general-purpose and user-friendly Python library that provides a rich set of functions and support for various image formats, making it an essential tool for developers to work with images in their projects.
It supports opening, manipulating, and saving many different image file formats, and users can also perform basic operations on images such as cropping, resizing, rotating, and changing image colors.
Pillow also lets you add text and shapes over images, providing an easy way to annotate your visuals.
This library is also the image processing library used by torchvison. It is powerful and easy to use. It is recommended to use.
2、OpenCV (Open Source Computer Vision Library)
OpenCV is undoubtedly one of the most popular image processing libraries. It was originally developed by Intel Corporation and has been widely used in the field of computer vision. It supports countless algorithms related to computer vision and machine learning, which help to understand visual data and make insightful decisions. OpenCV is also highly optimized for real-time applications, making it an excellent choice for video surveillance, self-driving cars, and advanced robotics.
OpenCV has the most functions and is faster than Pillow in terms of processing speed, so it is recommended to use it when speed is required.
Another point is that the channel read by OpenCV is BGR, while other libraries are RGB, so if you mix them, you need to convert, remember this code:
cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
3、Mahotas
Mahotas includes a set of functions for image processing and computer vision that is mostly done in high-performance C++ and uses multi-threading, making it very fast.
It also includes various morphological operations such as erosion, dilation and connected component analysis. These operations are the basis for tasks such as image binarization, denoising, and shape analysis. OpenCV has these functions, but Mahotas focuses more on image processing, and does not have everything like OpenCV. Therefore, Mahotas' API is simpler and more friendly. And it is easier to learn than OpenCV, but the speed is about the same.
4、Scikit-Image
Scikit-Image builds on the extended capabilities of the Scikit-Learn machine learning library, including more advanced image processing capabilities. So if you are already using Scikit for ML, then consider using this library.
It provides a complete set of image processing algorithms. It supports image segmentation, geometric transformations, color space manipulation and filtering.
Unlike many other libraries, Scikit-Image supports multidimensional images, which is helpful for tasks involving video or medical imaging. Scikit-Image integrates seamlessly with other Python scientific libraries such as NumPy and SciPy.
5、TensorFlow Image
TensorFlow Image is a TensorFlow module that supports image decoding, encoding, cropping, resizing, and conversion. You can also take advantage of TensorFlow's GPU support for faster image processing for larger datasets.
That is to say, if you use TF, you can use it as part of the training pipeline.
6、PyTorch Vision
Similar to TensorFlow Image, PyTorch Vision is part of the PyTorch ecosystem primarily for machine learning tasks related to image processing.
7、SimpleCV
Built on top of OpenCV, PIL (Python Imaging Library) and NumPy, SimpleCV provides users with a simple yet powerful set of functions and tools for loading, processing and analyzing images.
The design goal of SimpleCV is to make computer vision technology more reliable and easy to use for beginners and non-professionals. It provides a simple API that hides the underlying complexity, enabling users to quickly implement common computer vision tasks.
However, there is currently less official maintenance, so this project is likely to die.
8、Imageio
Imageio is a Python library for reading and writing a variety of image formats. It provides a simple yet powerful API that enables users to easily manipulate image and video data. Imageio provides a common data model that enables users to store image data in a variety of ways. It can represent image data using NumPy arrays, PIL image objects, or simple Python byte strings. And it provides the function of reading and writing video files frame by frame, which is very useful for processing video streams or extracting frames from videos.
9、albumentations
Albumentations is a Python library for image augmentation and data augmentation. It focuses on providing efficient, flexible and easy-to-use data augmentation methods in machine learning and computer vision tasks.
I have always regarded this library as an alternative to torchvision, because it not only has many data enhancement methods, but also can directly handle the enhancement of mask bbox.
10、timm
Timm is a PyTorch model library. Although it may not have anything to do with image processing, it provides a wide collection of pre-trained models and computer vision models, which is very helpful for us to do deep learning. Now it is a sub-project of huggingface, which means that the project has funding support, so there is no need to worry about development.
Summarize
Whether you are just getting started with basic image processing or exploring advanced machine learning models, these libraries provide the necessary tools for a wide range of image processing tasks.
https://avoid.overfit.cn/post/8912eb7ad4d04f359bbf802d62248bfa