Random subwindows and extremely randomized trees for image classification in cell biology

Abstract

Background: With the improvements in biosensors and high-throughput image acquisition technologies, life science laboratories are able to perform an increasing number of experiments that involve the generation of a large amount of images at different imaging modalities/scales. It stresses the need for computer vision methods that automate image classification tasks.

Towards generic image classification using tree-based learning: An extensive empirical study

“…Following and extending previous works (Marée et al, 2003(Marée et al, , 2004(Marée et al, , 2005(Marée et al, , 2007, we consider the generic problem of supervised image classification without any preconception about image classes, ie. it encompasses the recognition of numerous types of images under various image acquisition conditions.…”

“…We introduced previously different random subwindow sampling schemes (Marée et al, 2003(Marée et al, , 2005(Marée et al, , 2007. Random subwindows are square patches of random sizes extracted at random positions within images.…”

“…Default tests are made using a total of N ls = 1 million training subwindows (previous works (Marée et al, 2003(Marée et al, , 2005(Marée et al, , 2007 used only one hundred thousand subwindows) while a few others more intensive tests are performed with up to 50 millions subwindows. For a given dataset, the same number of subwindows are randomly drawn from each image, it equals N ls /N img where N img is the number of training images.…”

The One-Shot similarity kernel

“…An SVM classifier is then used to determine if the two images match by aggregating over all selected patches the output of pre-trained random decision trees. A similar method was used also for image classification in other domains [26,27].…”

Database of queryable gene expression patterns for Xenopus

“…One possibility is to rank images by pure graphical content similarity, which would presumably yield useful results if staining and imaging conditions were very carefully controlled. Algorithms have been developed to achieve this; for example, see Maree et al (2007) and Muller et al (2004) for a recent review. However, although such queries may be quite efficient, this does not constitute annotation, and will probably not generalize over images from different projects.…”