3.2Information Media (Compression, Encoding, Image/Audio/Video, VR/AR/CG)

Multimedia data such as images, audio, and video is very large in its raw form, so choosing the right compression method for the purpose is essential in practice. The FE exam does not simply test format-name memorization; it centers on judgments like "which compression method fits this use case" and "can some quality loss be tolerated." Understanding how analog audio and video are converted into digital data is likewise an important exam area.

3.2.1Lossless/lossy compression and key formats

• Lossless compression can fully restore the original data from the compressed data (e.g., ZIP, PNG); it is used for data like text or programs where not even a single bit of loss is acceptable. Lossy compression discards information that is hard for humans to perceive, achieving a higher compression ratio (e.g., JPEG, MP3); full restoration of the original data is not possible.
• JPEG is a lossy format suited to photographs and images with many colors and smooth gradations (raising the compression ratio introduces artifacts like blocky noise). PNG is a lossless format suited to images with sharp edges and transparency, such as logos and icons (it tends to produce larger files than JPEG for the same content). The typical exam pattern is choosing JPEG for photos, PNG for diagrams/logos.
• MPEG is a family of standards for compressing video (image plus audio), achieving a high compression ratio through inter-frame compression, which records only the differences between frames. MP3 is a lossy audio compression format that uses a psychoacoustic model to discard frequency components that are hard for the human ear to perceive.

3.2.2Sampling, quantization, encoding, and VR/AR/CG

• Converting an analog signal to digital data follows three stages: sampling—extract the signal's value at regular intervals (a higher sampling frequency more faithfully captures the original waveform) -> quantization—round each extracted value to a discrete step (more quantization bits allow smoother representation) -> encoding—convert the quantized value into a bit string in binary. Questions testing this order (sampling -> quantization -> encoding) are classic.
• VR (virtual reality) immerses the user in a fully constructed virtual space via a headset or similar device (the real-world view is blocked out). AR (augmented reality) adds virtual information on top of the real-world view, such as overlaying data on a smartphone camera feed (the real-world view remains visible). The deciding question for telling VR and AR apart is whether reality is replaced or overlaid.
• CG (computer graphics) broadly covers techniques for generating and processing images with a computer, ranging from 2D illustration to 3D modeling and animation. It is applied widely, from film and games to product-design simulation.

Consider posting a product catalog on an in-house system. Actual product photos have smooth, numerous color gradations, so compressing them with JPEG keeps file size well down—but pushing the compression ratio too high introduces visible blocky artifacts, so a setting must balance image quality against file size. In contrast, for a company logo or an operation icon with sharp edges, few colors, and a transparent background, you should use lossless PNG rather than lossy JPEG, which would blur the edges. Jumping to the conclusion that "higher compression ratio is always better" here leads to a logo that looks blurred when saved as JPEG. Next, consider recording and digitizing voice guidance for internal training: you first take an analog sound wave arriving from a microphone and cut out its value at regular intervals via sampling (e.g., 44,100 times per second at 44,100 Hz); you then round each extracted value to a predetermined number of steps via quantization (e.g., 65,536 steps at 16 bits); and finally you turn that number into a binary bit string via encoding, at which point the digital audio data is complete. Keep in mind the tradeoff that raising the sampling frequency brings the result closer to the original sound, and raising the quantization bit depth makes it smoother, but both increase the data volume. If you want to help warehouse pickers by overlaying the location of the next item to grab onto what they actually see through goggles, AR—which keeps the real view intact while adding information—is the right fit. Conversely, if you want to train workers on operating dangerous machinery in a safe environment, VR—which fully blocks out reality and immerses the user in a constructed virtual space—is the right fit.

3.2.3Section summary

• Lossless = fully restorable (PNG, etc.); lossy = discards imperceptible info for a higher ratio (JPEG, MP3, etc.). Photos = JPEG, logos = PNG is typical
• Digitization order: sampling (time axis) -> quantization (amplitude) -> encoding (to binary). Higher frequency/bit depth is more faithful but increases data volume
• VR = replaces reality for immersion; AR = overlays information onto reality. CG broadly covers generating/processing images