The raw image data from the Acoustic Daylight Ocean Noise Imaging System (ADONIS) consist of only 126 pixels, arranged in an elliptical shape of major and minor axes 14×11 pixels. An estimator is presented which produces higher spatial resolution from these raw images using a maximum-likelihood estimator (MLE) coupled with a point-spread function model of the imaging system. The technique converges iteratively on a constrained MLE of the intensity distribution in each image frame. The MLE performs better than traditional interpolation methods. The estimator is also extended to constructing improved images from a sequence of image frames, from which the relative values and convergence rates of different frequency components becomes apparent. Examples of the technique are presented for both simulated and experimental data taken from a deployment of the ADONIS system. [Work funded by NUS and DSO, Singapore. ADONIS data provided by Scripps Institution of Oceanography.]