Section: New Results

Line detection in microarray scanner images

Participants : Alice Bergonzoni, Charles Kervrann.

In this study, we have studied two approaches to automatically detect straight lines in images (tool-slide) for calibrating scanners designed by Innopsys company. The Hough transform has been investigated and is able to produce satisfying results provided the algorithm parameters are carefully adjusted (see Fig. 12 ). To overcome this difficulty, we have evaluated the potential of a contrario approach [28] which is well ground theoretically and requires no object prior and parameter adjustment. According to the Helmholtz principle which is based on the a contrario approach, any structure is considered in a white noise image as a deviation from randomness. A meaningful segment is detected when the expectation of its number of occurrences in a white noise image (i.e. number of false alarms) is low. We have evaluated the potential of this method and performed experiments using the LSD algorithm [46] inspired from [28] .

Partners:: V. Paveau (Innopsys)

Figure 12. Detection of lines in an image (tool-slide) (pixel size: $3\mu \text{m}\times 3\mu \text{m}$).