A real-time method for removing stripe artifacts from infrared-reflectance (IR) images

Abstract

Purpose : Real-time tracking based on IR images from a broad-line fundus imager requires that the images be relatively free of stripe artifacts, which can obscure retinal features. We propose a real-time, frequency-domain algorithm to remove these stripe artifacts.

Methods : IR preview images are created by acquiring overlapped thin stripes, resulting in a 2-D image with stripe artifacts. The accuracy of any retinal tracking algorithm would suffer from the presence of these stripe artifacts in images. It is essential to remove the stripe artifacts prior to any image processing that is needed for tracking. Our real-time stripe removal algorithm (implemented via Intel IPP library) is based on spatial frequency-domain filtering. Basically, the horizontal periodic structures are removed by notching out frequencies associated with stripes pattern and leaving other frequencies of the fast Fourier transform (FFT) relatively unchanged (Fig 1).

Performance of the algorithm was evaluated using 100 IR images of different quality from a CLARUS™ 500 (ZEISS, Dublin, CA). The size of images is 768×624 pixels (width x height) covering a retinal area of approx 11.5 × 9.4 mm. A grader evaluated the results of the algorithm by grading as “acceptable” (stripes removed) or “unacceptable” (stripes still present).

Results : Fig 2 shows two examples before and after stripe removal along with the magnitude of gradient for each example. The magnitude of gradient before stripe removal shows enhanced edges of horizontal artifacts which affect the performance of retinal tracking. The algorithm removed the stripes of all 100 IR images successfully as reported by the grader. The execution time of the algorithm measured 1 ms on average using Intel i7-8850H CPU, 2.6 GHz, 32 GB RAM.

Conclusions : We developed a real-time method for removing stripe artifacts from IR images. The algorithm successfully removed the stripes from all test images. The removal of stripe artifacts improves clarity of retinal structures, essential for feature-based retinal tracking algorithms.

This is a 2020 Imaging in the Eye Conference abstract.

 

Fig 1: Stripe removal algorithm from left to right: IR image with stripes, FFT of IR image, notch-filtering by removing frequencies associated with stripes pattern, IR image without stripes after inverse FFT.

 

Fig 2: Two examples of IR CLARUS images (left column) including their magnitude of gradients (right column) before and after stripe artifact removal (top and bottom row, respectively).