In the last few years, autostereoscopy has become an emerging technology. This technique uses n acquisitions from the same scene and introduces therefore a new data redundancy dimension. This process generates a large amount of data (typically n times more than a single image) that needs to be compressed for further network applications. It must be an almost-lossless scheme since autostereoscopy is very sensitive to artifacts. Thus common JPEG compression is not suitable for this application. A simple way to compress an image sequence is to take each view and compress it separately with well-known near-lossless algorithms like JPEG at high quality, JPEG2000 or JPEG-LS. This approach is very easy to implement but does not reduce the inter-view redundancy and can be improved by considering the whole image set. In this paper, we present an alternative to traditionnal methods used for image compression: MICA (Multiview Image Compression Algorithm). MICA is a near-lossless scheme that exploits the positive-sided geometric distribution (PSGD) of pixels from the difference of two consecutive views with a modified arithmetic coding. However, we choose to keep a lossless compression scheme (JPEG-LS) for two specific views in order to avoid error propagation during the decoding process. This algorithm has a low complexity, and can be easily parallelized either on CPU or on GPU for real-time applications or autostereoscopic videos.