Interactive Segmentation and Classification Library using Machine Learning and Geometric Features

Active Segmentation aims of providing a general purpose workbench that would allow biologists to access state-of-the-art techniques in machine learning and image processing to improve their image segmentation results. Currently, Active Segmenation have various geometric features like Laplace of Gaussian , Gaussian Derivatives etc. It also provide various machine learning algorithms like Support Vector machine, Naive Bayes etc available in WEKA library. The platform allows researchers to add their own feature extraction and machine learning algorithms.

The major goal of active segmentation is to provide a well documented tool for users along with visual insights so that user can understand the underlined methodology

An detailed description of each screen can be fould in UserGuide


Introduction

In order to use active segmentation, user need to get familiar with basic concepts of scale space theory and machine learning. Scale space is the the representation of image at continous scale and user can configure the scale using the filters user interface depending upon the problem.

  • Filters
    The user can select or deselect filters depending on their problem. The active segmentation provide default list of filters but users can add their customer filters. The user can even tune the parameters of filters using GUI.

  • Feature Extraction
    Researchers or users need to mark few samples of interest for each of the probable class. The most common example is two class problem when user want to segment cell from background. It is an iterative processing and user can add new samples in each iteration in order to improve the segmentation results.

  • Learning
    Learners are the models that try to learn the function that maps feature space to classes space. The default learner in active segmentation is support vector machines. The user can change the learner and even modify the parameters of the learner from the learning screen

  • Evaluation
    We computed various evaluation metric like precision, Recall, Area under ROC curve and Area under PR curve in order to evaluate the robustness of the machine learning models

Citing

Paper link will be provided soon