Lossy Light Field Compression

The aim of this paper is to propose, implement and evaluate a new lossy compression method for light field images. The proposed method extends the JPEG method to four dimensions and brings new ideas which lead to better compression performance. Correlation between light field views is exploited in both dimensions by four-dimensional discrete cosine transformation. The lossy part of the encoding is performed by quantization, similarly to the JPEG method. The proposed method is implemented as a program library in a C++ language. This paper compares the proposed method to JPEG, JPEG 2000 and HEVC intra image compression methods and HEVC video compression method. The results show that the proposed method outperforms the reference methods with images with a higher amount of views. HEVC video method is better for images with fewer views or for very low bitrates