Image Feature Extraction Software

Summary

The Feature Extraction Software is a flexible tool to extract and visualize a variety of image features from large images (Giga Bytes size). It supports over 30 distinct features that can be parametrized, which can be extracted from different user-defined regions of interest. It is implemented in the Java programming language and can run locally on a desktop or take advantage of parallel computing on a cluster (e.g. Hadoop). Figure 1 illustrates hexagon spatial regions partitioning stem cell colonies that are color-coded based on the extracted average intensity value.

Description

Different types of features are supported, such as:

Intensity (8 Features)

Mode, Mean, Standard Deviation, Skewness, Kurtosis, Fifth, Sixth moments, Entropy

Texture (15 Features)

Correlation, Energy, Contrast and Homogeneity, Texture Entropy and more

Spatial (10+ Features)

Area, Perimeter, Circularity, Orientation, Eccentricity, Distance from border, Centroid and more

Features can be extracted from different Regions Of Interests (ROI):

• Squared grid
• Hexagonal grid
• Arbitrary shapes by providing a mask (e.g. colonies)
• Masks can be sub-partitioned into squares or hexagons of a given size

Figure 2 illustrates the options for extracting image features with or without a mask, and with other options with a mask that enable feature extractions over various regions of interest.

The extracted features are outputted by default as CSV tables. When extracting features on a number of different images (e.g. evolution of colonies over time), the CSV tables can be merged into a single sqlite database, facilitating data-mining on the dataset.