This Invited Paper presents the 25 years history and the future prospect of JPEG 2000 standards. JPEG 2000 standards (ISO/IEC 15444 series) have been produced by ISO/IEC JTC 1/SC29/WG 1 and common texts are also registered as ITU-T Recommendations (ITU-T T.800 series). JPEG 2000 Part 1 was born in year 2000 and from then around 15 parts have been standardized to compose JPEG 2000 family of standards. Main components of JPEG 2000 are DWT (Discrete Wavelet Transform), EBCOT (Embedded Block Coding with Optimal Truncation) and MQ coder. In JPEG 2000, a lot of functions for image built-up, image exhibition, image processing are integrated in the coding system and various extensions needed for specific applications are provided in the family standards. The coding performance is highly increased from original JPEG standard and JPEG2000 has been adopted in various fields such as Digital Cinema, Document Archives around the world, and PDF specifications. The recent update is the introduction of High-Throughput JPEG 2000 (HTJ2K) and the throughput is improved by an order of magnitude at the expense of slightly reduced coding efficiency. It is expected that the birth of HTJ2K would strongly accelerate the penetration of JPEG 2000 in the near future.

Omnidirectional (360°) cameras are being increasingly used for recording and presenting a wide range of scenes such as street views, dashboard cameras, real estate literature introductions, and drone imagery. As most of these scenes are captured using fisheye lenses with large distortion, omnidirectional cameras are not commonly used for three-dimensional shape measurement or image processing. Although various conventional methods have addressed these limitations by applying perspective projection transformation to 360° images, this transformation increases the processing time. A fundamental challenge in image recognition is the detection of straight and parallel lines. Accordingly, I propose a method to detect these lines directly from 360° images without using perspective projection transformation and then estimate the axial direction and camera attitude of the detected parallel lines. Through experiments, I compared the proposed method with conventional line detection using perspective projection and demonstrated a short processing time of the proposed method and accurate detection of parallel lines. The processing times of the proposed method were approximately 10 and 30 ms per frame for cameras with 2K and 4K resolutions, respectively. In addition, the detection accuracy of parallel lines was 87% when excluding environment factors. The attitude estimation errors were approximately 11° and 5° for 2K and 4K resolutions, respectively.