This paper investigates the lossless compressibility of the images along a typical processing chain of a plenoptic image, including the stages of deomosaicing, devignetting, alignment, and rectification. The processing chain is deterministic, having at the input only the captured sensor image and a set of camera parameters specific to the particular focus and zoom settings used when taking the picture, while at the end it has the rectified set of image views, and hence the lossless coding of any intermediate image may ensure that the decoder can obtain the same final rectified images. However, some of the processing steps are introducing irreversible losses, and hence the initial input sensor image is not obtainable from some of the (later) intermediate images. We study the compressibility of the images at the main processing steps first using general publicly available compressors and then when using the redundancies specific to plenoptic images, in order to shed light into the practical aspects of image compression of plenoptic images, and also to gain intuition on the changing of information content along the processing chain. Recently there was a wide interest in studying the compressibility of the images at different points along a typical processing chain of a plenoptic image, including the stages of deomosaicing, devignetting, alignment and rectification. The typical processing chain [1] has at the input the captured sensor image, an additional calibration (white) image, and a set of camera parameters specific to the particular focus and zoom settings used when taking the picture, while at the end it has the rectified set of image views. The processing along the chain is not standardized and we use here [1], which was also recommended in a recent lossy compression challenge proposal at ICME 2016. Some of the processing steps are introducing irreversible losses, and may include information or additional particular transformations that are not always available at the decoder. The images at different points in the chain have different sizes and different statistical properties and hence their compressibility is different, see the figure. Encoding the image B at the start of the rectification chain will require that the decoder performs the rectification at its side, while encoding the rectified image LF will reduce the tasks of the decoder and the need of additional information required for performing the rectification at the decoder. We presented in [2] a sparse modelling approach for predictive compression in the case when the plenoptic image is available in its form LF, which consists of 15 × 15 side views. We review here the results of [2] and discuss the applicability of the scheme to the other points in the processing chain.