The impact of different stress factors on plant growth can be assessed through the measurement of sprout length. Measuring the sprout length can be done through traditional methods (e.g. use of calipers) or through image analysis. The use of ImageJ, an image analysis program developed at National Institute of Health, is an established validated way to do the measurement (Schneider et al., 2012). But the line segmentation measure feature to measure the sprouts is very inefficient and time consuming when dealing with hundreds of samples. In this work we developed a computer vision-based approach, SproutCV, to automate the measurement process. After image preprocessing, where mean shift filtering, grayscale, Gaussian blur, and morphological operations were applied, Otsu’s thresholding was implemented to identify sprout regions. We then skeletonized the output image and converted it to a network graph, wherein each pixel in an 8-connectivity is the node and the Euclidean distance between each pixel is the weighted edge. The skeleton lines were simplified further using Douglas-Peucker algorithm to remove unnecessary pixels. Dijkstra’s algorithm was utilized to find the two farthest nodes in the graph and the shortest path between them. The length of this final graph path was measured, and this represented the sprout length. SproutCV achieved a coefficient of determination (R2) of 0.9991 for 566 sprouts in comparison to measurements done in ImageJ, with a mean absolute percent error of 1.74%. This shows that SproutCV is an effective and accurate way of automating measurements of sprouts. Future improvements to this algorithm can expand its applications on automating measurements of various objects, from red blood cells in medical imaging to fishes in marine biology.

More information here.