Texture Analysis

Overview

MSc Project description

This project explores how dictionary methods can be used to describe textures in histology images of mouse brains. We aim to investigate the following problems:

1. Can we use a convolutional neural network (CNN) to learn better texture descriptors ?

2. Can we develop a pursuit method similar to Independent Component Analysis (ICA) that finds the salient textures in the image ?

Diaries

• Cell-based Method
• Deep Neural Network
• Histogram-based Method

References

• MICCAI 2015 Survey
• Papers on Cell Detection
• Advice from Prof. Zhuowen Tu

Boosting

• Tutorials on boosting: http://www.boosting.org/tutorials
• Boosting Neural Networks:
  • http://www.iro.umontreal.ca/~lisa/pointeurs/ada-nc.pdf
  • Got to http://www.boosting.org/publications and search for "neural networks"
• Googling "boosting sift descriptors" yields:
  • Weak hypotheses and boosting for generic object detection and recognition
  • CHoG: Compressed Histogram of Gradients
  • Improvements of Object Detection Using Boosted Histograms

Students

• Yuncong Chen
• Jiaxu Zhu
• Ruogu Liu

Data

• Images: /oasis/projects/nsf/csd395/yuncong/CSHL_data_processed/<stack>_lossless_aligned_cropped

Available stacks are MD589, MD592, MD593, MD594, MD595

Each stack has ~130 sections (dimension 15,000 x 10,000 RGB, tif, ~400MB)

• There are two other versions of the same images:
  • <stack>_lossless_aligned_cropped_grayscale: images converted to grayscale
  • <stack>_lossless_aligned_cropped_downscaled: compressed with lossy jpeg, ~40MB each, for visualization purposes

• Masks: /oasis/projects/nsf/csd395/yuncong/CSHL_data_processed/<stack>_thumbnail_aligned_cropped_mask

contains masks for the thumbnails. To use them on the original-sized images, scale them by 32 on each dimension.