A complete guide to how image data flows through vision systems, from CMOS sensors to AI inference and video encoding.Target AudienceCamera development engineersSurveillance system developersRobotic vision engineersAI vision algorithm engineersEmbedded software engineersDevelopers working on HiSilicon, Rockchip, Ambarella, and NVIDIA Jetson platformsTable of ContentsWhy Are There So Many Image Formats?End-to-End Image Processing Pipeline OverviewRAW: The Unprocessed Sensor OutputISP: The Core of Image QualityRGB: The Most Intuitive Image FormatYUV: The Backbone of the Video IndustryDifferences Between YUV444, YUV422, and YUV420YUYV: The Most Common Format for USB CamerasNV12: The De Facto Standard for Video ApplicationsNV21 and I420MJPEG: Compressed Image StreamsWhat Are H.264 and H.265?Image Formats in AI VisionData Size Comparison of Common FormatsFormat Selection in Engineering PracticeCommon QuestionsSummary1. Why Are There So Many Image Formats?When first working with vision systems, you’ll encounter a long list of terms: RAW10, RAW12, RGB888, RGB565, BGR888, YUYV, NV12, NV21, I420, MJPEG, H.264, H.265. A common question arises: Why do we need so many formats for an image?These formats are not competing standards—they correspond to different stages of the image processing pipeline:Capture → Processing → Display → Encoding → TransmissionEach stage requires a format optimized for its specific task.