Object detection

ModularImageAnalysis (MIA) is an ImageJ plugin which provides a modular framework for assembling image and object analysis workflows. Detected objects can be transformed, filtered, measured and related. Analysis workflows are batch-enabled by default, allowing easy processing of high-content datasets.

MIA is designed for “out-of-the-box” compatibility with spatially-calibrated 5D images, yielding measurements in both pixel and physical units.  Functionality can be extended both internally, via integration with SciJava’s scripting interface, and externally, with Java modules that extend the MIA framework. Both have full access to all objects and images in the analysis workspace.

Workflows are, by default, compatible with batch processing multiple files within a single folder. Thanks to Bio-Formats, MIA has native support for multi-series image formats such as Leica .lif and Nikon .nd2.

Workflows can be automated from initial image loading through processing, object detection, measurement extraction, visualisation, and data exporting. MIA includes near 200 modules integrated with key ImageJ plugins such as Bio-Formats, TrackMate and Weka Trainable Segmentation.

Module(s) can be turned on/off dynamically in response to factors such as availability of images and objects, user inputs and measurement-based filters. Switches can also be added to “processing view” for easy workflow control.

MIA is developed in the Wolfson Bioimaging Facility at the University of Bristol.

QuantiFish is a quantification program intended for measuring fluorescence in images of zebrafish, although use with images of other specimens is possible. This package is geared towards analysis of fluorescent infection models. The software is designed to automate processing of images of single fish, and outputs results as a .csv file. Alongside measures of total fluorescence above a threshold, this package also introduces several measures for dissemination and distribution of fluorescence throughout the specimen.

Multi-template matching can be used to localize multiple objects using one or a set of template images.

Contrary to previous implementations that allow to use only one template, here a set of templates can be used or the initial template(s) can be transformed by rotation/flipping.

Multiple objects detection without redundant detections is possible thanks to a Non-Maxima Supression relying on the degree of overlap between detections.

The solution is available as a Fiji plugin (Multi-Template Matching AND IJ-OpenCV update sites), as a Python package (Multi-Template-Matching on PyPI) and as a KNIME workflow (via KNIME Hub).

This is an implementation of Mask R-CNN on Python 3, Keras, and TensorFlow. The model generates bounding boxes and segmentation masks for each instance of an object in the image. It's based on Feature Pyramid Network (FPN) and a ResNet101 backbone.