Hyperspectral imaging has become an interesting non-destructive way to spatially characterize various materials such as vegetation. Because several leaf biochemical compounds affect the visible and near-infrared reflectance, they can potentially be retrieved from the measured spectral profile. At leaf scale, usual methods are either statistically- or physically-based. On the one hand, statistical methods aim at extracting a statistical relationship between spectral data and targeted variable(s), thus potentially requiring the construction of a large calibration database to get to an acceptable robustness. On the other hand, physically-based methods use the knowledge of leaf optical properties to retrieve the targeted variables without any calibration. Although being highly effective in ideal laboratory conditions, these methods do not scale well to the close range remote sensing case because they require the signal to be measured in every direction using an integrating sphere. In this study, we propose an improvement of a physically -based model that overcomes the need for multi-angle measurements. This model is then applied to the retrieval of water and dry matter contents from NIR hyperspectral images. This method is based on the PROSPECT radiative transfer model that has been much used to characterize the relationship between the directional-hemispherical reflectance of monocotyledon and dicotyledon species and their biochemical content at leaf level. However, because PROSPECT simulates the sum of both specular and diffuse reflected fluxes over the whole hemisphere, it cannot directly be applied to hyperspectral images of leaves. Using a single light source assumed to be directional, we show that adding two parameters to the PROSPECT model enables the model to be used with hyperspectral measurements. The first parameter is a multiplicative term that is related to local leaf angle and illumination zenith angle. The second parameter is an additive specular-related term that models surface bidirectional effects. The resulting enhanced PROSPECT-based model can therefore be inverted from hyperspectral measurements to retrieve the foliar content, surface effects and leaf local topography. We tested this method on seven leaf species with images acquired in laboratory using a SWIR (1-2.5 µm) hyperspectral camera. Each leaf was imaged in three positions corresponding to average incident angles of 10°, 30° and 50°. Lighting was provided by a collimated halogen source positioned close to the camera. After each image acquisition, water and dry matter contents were evaluated using destructive measurements as these PROSPECT variables have the greatest influence on reflectance in this spectral range. When tested on horizontal leaves, our model accurately retrieved water (R²=0.98,RMSE=24%) and dry matter (R²=0.85,RMSE=16%) contents. Results obtained with both flat and tilted leaves did not show significant decrease in performance, either for water (R²=0.95,RMSE=28%) or dry matter (R²=0.84,RMSE=17%) contents. These results were better than those obtained with PROSPECT alone, i.e., R²=0.79 and RMSE=28% for water content, and R²=0.67 and RMSE=26% for dry matter content. In addition, the estimated spatial distribution of model parameters was clearly not affected by variable surface effects and leaf local topography, thus proving the relevance of the proposed model for hyperspectral imaging compared to original PROSPECT.