Computer graphics (CG or just graphics) is a field of computer science that deals with visual information. The field doesn't have strict boundaries and can blend and overlap with other possibly separate topics such as physics simulations, multimedia and machine learning. It usually deals with creating or analyzing 2D and 3D images and as such CG is used in data visualization, game development, virtual reality, optical character recognition and even astrophysics or medicine.
We can divide computer graphics in different ways, traditionally e.g.:
Since the 90s computers started using a dedicated hardware to accelerate graphics: so called graphics processing units (GPUs). These have allowed rendering of high quality images in high FPS, and due to the entertainment and media industry (especially gaming), GPUs have been pushed towards greater performance each year. Nowadays they are one of the most consumerist hardware, also due to the emergence of general purpose computations being moved to GPUs (GPGPU), lately especially mining of cryptocurrencies and training of AI. Most lazy programs dealing with graphics nowadays simply expect and require a GPU, which creates a bad dependency and bloat. At LRS we try to prefer the suckless software rendering, i.e. rendering on the CPU, without GPU, or at least offer this as an option in case GPU isn't available. This many times leads us towards the adventure of using old and forgotten algorithms used in times before GPUs.
This is a general overview of 3D graphics, for more technical overview of 3D rendering see its own article.
3D graphics is a big part of CG but is a lot more complicated than 2D. It tries to achieve realism through the use of perspective, i.e. looking at least a bit like what we see in the real world. 3D graphics can very often bee seen as simulating the behavior of light; there exists so called rendering equation that describes how light behaves ideally, and 3D computer graphics tries to approximate the solutions of this equation, i.e. the idea is to use math and physics to describe real-life behavior of light and then simulate this model to literally create "virtual photos". The theory of realistic rendering is centered around the rendering equation and achieving global illumination (accurately computing the interaction of light not just in small parts of space but in the scene as a whole) -- studying this requires basic knowledge of radiometry and photometry (fields that define various measures and units related to light such as radiance, radiant intensity etc.).
In 2010s mainstream 3D graphics started to employ so called physically based rendering (PBR) that tries to yet more use physically correct models of materials (e.g. physically measured BRDFs of various materials) to achieve higher photorealism. This is in contrast to simpler (both mathematically and computationally), more empirical models (such as a single texture + phong lighting) used in earlier 3D graphics.
Because 3D is not very easy (for example rotations are pretty complicated), there exist many 3D engines and libraries that you'll probably want to use. These engines/libraries work on different levels of abstraction: the lowest ones, such as OpenGL and Vulkan, offer a portable API for communicating with the GPU that lets you quickly draw triangles and write small programs that run in parallel on the GPU -- so called shaders. The higher level, such as OpenSceneGraph, work with abstraction such as that of a virtual camera and virtual scene into which we place specific 3D objects such as models and lights (the scene is many times represented as a hierarchical graph of objects that can be "attached" to other objects, so called scene graph).
There is a tiny suckless/LRS library for real-time 3D: small3dlib. It uses software rendering (no GPU) and can be used for simple 3D programs that can run even on low-spec embedded devices. TinyGL is a similar software-rendering library that implements a subset of OpenGL.
Real-time 3D typically uses an object-order rendering, i.e. iterating over objects in the scene and drawing them onto the screen (i.e. we draw object by object). This is a fast approach but has disadvantages such as (usually) needing a memory inefficient z-buffer to not overwrite closer objects with more distant ones. It is also pretty difficult to implement effects such as shadows or reflections in object-order rendering. The 3D models used in real-time 3D are practically always made of triangles (or other polygons) because the established GPU pipelines work on the principle of drawing polygons.
Offline rendering (non-real-time, e.g. 3D movies) on the other hand mostly uses image-order algorithms which go pixel by pixel and for each one determine what color the pixel should have. This is basically done by casting a ray from the camera's position through the "pixel" position and calculating which objects in the scene get hit by the ray; this then determines the color of the pixel. This more accurately models how rays of light behave in real life (even though in real life the rays go the opposite way: from lights to the camera, but this is extremely inefficient to simulate). The advantage of this process is a much higher realism and the implementation simplicity of many effects like shadows, reflections and refractions, and also the possibility of having other than polygonal 3D models (in fact smooth, mathematically described shapes are normally much easier to check ray intersections with). Algorithms in this category include ray tracing or path tracing. In recent years we've seen these methods brought, in a limited way, to real-time graphics on the high end GPUs.
All content available under CC0 1.0 (public domain). Send comments and corrections to drummyfish at disroot dot org.