Most of European fish stocks are assessed using age-based models, and otolith interpretation for age estimations costs several million euros annually. In this context, automated ageing systems would provide a mean to 1) standardize ageing, 2) control ageing consistency within and between ageing laboratories 3) build interpreted image data bases ensuring the information conservation and sharing and 4) improve growth studies while reducing the cost of the acquisition of age data. This paper presents a review of different image-based tools for automatic fish ageing from otolith features within the framework of AFISA (EU STREP project on Automated FISh Ageing coordinated by Ifremer). The full automatic ageing process can be divided into three main steps: 1) image acquisition and pre-processing, 2) information extraction and representation from otolith images, and 3) age estimation. To illustrate what is currently feasible, some algorithms are tested on two image databases of North Sea cod and Eastern Channel plaice otoliths. Acquisition and pre-processing step Currently, otolith images are acquired manually one after one which takes about one minute per image. An automatic system can potentially spares a lot of time during this part of the process, hence reducing also the cost of acquisition. Here, we propose an automatic acquisition method where otoliths are first placed on a grid, and then a camera coupled with a motorized microscope acquires images of the whole grid. An algorithm based on a morphological approach is applied on the resulting image mosaic to automatically detect the position of each otolith. Finally, the individual otolith images are extracted and saved in a database. Otolith images depict clearly oriented ring structures but also contain artefacts that can penalize the treatments. During the pre-processing step, a filter can be used to removethe artefacts. As ring width is highly variable, classical isotropic Gaussian filter can damage the information while failing to remove artefacts. We thus propose an anisotropic filter with adapted orientation and scale that permit to follow the specific geometry of otolith structures. Extraction of 2D information Many features can be extracted from otolith images such as growth axes, growth marks (eg seasonal) and morphological information. Classical methods have been mainly based on 1D information. Previous 1D approaches required a manual definition of a growth axis on which rings are then detected. Such methods are sensitive to noise and suffer from a lack of standardization. Compared to the 1D approach, the 2D methods reduce the noise sensitivity by integrating all the information available in the image and thus reduce the negative impact of local artefacts on final estimation. Moreover, compare to manuals methods, the automation allow standardizing the way information is acquired, so that comparisons between otoliths are made more relevant, even when they have been acquired by different operators in different laboratories. Finally, recent advances allow extracting curvilinear growth axes from the core to the edge, using an estimation of the local orientation of structures. For the extraction of growth marks like seasonal macrostructures, the classical 1D methods use the 1D opacity signal along the growth axis. To reduce the noise sensitivity, some methods have proposed to integrate the information available inside an angular section around the growth axis. We propose here to consider the whole otolith image as a 2D signal that contains growth marks. The use of 2D methods not only reduces the sensitivity to noise, but also gives access to the morphogenesis and provides new metrics (eg local growth and opacity anisotropy). Another advantage of the 2D is that the information can be represented in a more convenient way to extract interesting features more precisely. For example, the 2D contrast-invariant representation attenuates the variations of opacity coming from the conditions of acquisition which results in highlighted variations of opacity that are induced by environmental factors (eg temperature, food). Analyse and interpretation For age estimation, classical methods of analyse are just basically based on checking and counting the detected seasonal marks. A more reliable estimation can be reach using more advanced methods like statistical learning. Given some known examples, these learning methods are able to generalise a law to estimate the age not only based on the number of detected marks but also using other features like the seasonal macrostructures positions, shape and growth axis. Conclusion Recent advances in computer vision gives more reliable methods to extract information from otoliths and to interpret these features to estimate the age and growth of fish. But these methods should not be seen as being able to fully substitute to expert excepted in trivial cases. They should rather been seen as tools to provide automatically extracted information that requires a subsequent verification by the expert both for age and growth estimation.