Surveillance Image Compression Based on ROI and Image Restoration

Common compression techniques and standards are suitable for general purpose applications and are less effective for specific applications. This paper proposes an application-specific compression technique for traffic surveillance images, which will fully take advantage of the characteristics of a target application and maximally reduce the redundancy information. The compression algorithm treats the car plate, foreground targets and the background as three different regions, which are located by car plate detection algorithm and background subtraction, with decreased interest and processes with different compression ratio and precision. A specific filter will be used in decompression to reduce combination difference on the restored image. Compared with standard JPEG compression, this multi-level compression algorithm can save more than 60 percents of the storage space with extra 113 ms while maintaining fine vision effect. This algorithm can be extended to still image group compression that owns the same background.