We propose a learning retina encoder (RE) with tunable spatiotemporal functions and corresponding perception-based dialog procedure for a retina implant as visual prosthesis. A learning DSP-based RE was developed for real time implementation of about 200 individually tunable spatiotemporal receptive field (RF)-filters approximating M- or P-ganglion cells to map visual patterns onto pulse trains as stimulation inputs for implanted microcontact foils. An experimental environment providing visual feedback for humans with normal vision was developed to test the adaptive dialog module. With the presented tuning schema, interaction of RE and dialog module becomes possible even before tests with blind subjects.