Context. Direct imaging of circumstellar disks at high angular resolution is mandatory to provide morphological information that constrains their properties, in particular the spatial distribution of dust. For a long time, this challenging objective was, in most cases, only within the realm of space telescopes from the visible to the infrared. New techniques combining observing strategy and data processing now allow very high-contrast imaging with 8-m class ground-based telescopes (10 −4 to 10 −5 at ∼1 ) and complement space telescopes while improving angular resolution at near infrared wavelengths. Aims. We present the results of a program carried out at the VLT with NACO to image known debris disks with higher angular resolution in the near-infrared than ever before in order to study morphological properties and ultimately detect the signpost of planets. Methods. The observing method makes use of advanced techniques of adaptive optics, coronagraphy, and differential imaging, a combination designed to directly image exoplanets with the upcoming generation of "planet finders" such as GPI (Gemini Planet Imager) and SPHERE (Spectro-Polarimetric High contrast Exoplanet REsearch). Applied to extended objects such as circumstellar disks, the method is still successful but produces significant biases in terms of photometry and morphology. We developed a new model-matching procedure to correct for these biases and hence provide constraints on the morphology of debris disks. Results. From our program, we present new images of the disk around the star HD 32297 obtained in the H (1.6 μm) and Ks (2.2 μm) bands with an unprecedented angular resolution (∼65 mas). The images show an inclined thin disk detected at separations larger than 0.5−0.6 . The modeling stage confirms a very high inclination (i = 88 • ) and the presence of an inner cavity inside r 0 ≈ 110 AU. We also find that the spine (line of maximum intensity along the midplane) of the disk is curved, which we attribute to a large anisotropic scattering-factor (g ≈ 0.5, which is valid for an non-edge-on disk). Conclusions. Our modeling procedure is relevant to interpreting images of circumstellar disks observed with angular differential imaging. It allows us to both reduce the biases and estimate the disk parameters.