Purchase this article with an account.
N. Parikh, M. Humayun, J. Weiland; Image Processing Algorithm for Cueing Salient Regions Using a Digital Signal Processor for a Retinal Prosthesis. Invest. Ophthalmol. Vis. Sci. 2008;49(13):4048. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To develop an image processing algorithm capable of cueing retinal prosthesis patients to direct their gaze to important objects in the peripheral visual field. Such an algorithm may aid implant recipients in understanding unknown environments by directing them to look towards important areas.
Initial algorithm development is done using MATLAB and then implemented on a digital signal processor (DSP TMS320DM642, 720 MHz Imaging Developers Kit, Texas Instruments, Inc.). The algorithm uses the saturation and intensity values from the HSI color space along with the high pass information of an image and constructs feature maps from these three information streams. Feature maps undergo a normalization process and are combined to form a final saliency map from which salient regions are detected. The algorithm is loosely based on Itti’s model of primate visual attention (primate model), with several key differences. Our algorithm uses the last 4 levels of the pyramids as compared to 9 levels used by the primate algorithm. Our algorithm thus concentrates more on low frequency which leads to the detection of larger details than small and fine details. Gaussian pyramids are created at 9 levels by successive decimation and low pass filtering but only the last 4 levels are used in constructing feature maps. Only 3 information streams are used in our algorithm, versus 7 in the primate model.
Our algorithm implementation on the DSP operates at about 1-2 frames/sec. As a comparison, algorithms implementing only 1 of 7 information streams in the primate model run at less than 1 frame per second on the DM642 DSP. On demand by the user, one or more cues can be given in the decreasing order of saliency to the user. The user can then scan the region around the direction of the cue(s) instead of scanning the entire scene which can be more time consuming. Field of view for recipients of retinal prosthesis is limited to 20 degrees. The algorithm cues 8 regions (regions to the left, right, top, down, top-left, top-right, bottom-left and bottom-right) falling outside the field of view using visual or audio cues.
A saliency algorithm has been implemented to provide retinal prosthesis patients with cues to salient regions in a visual field. This algorithm may be helpful since retinal prostheses are generally limited to central visual field.
This PDF is available to Subscribers Only