Methodology for Increasing the Measurement Accuracy of Image Features

Summary

We present an optimization methodology for improving the measurement accuracy of image features for low signal to noise ratio (SNR) images. By superimposing known background noise with high quality images in various proportions, we produce a degraded image set spanning a range of SNRs with reference feature values established from the unmodified high quality images. We then experiment with a variety of image processing spatial filters applied to the degraded images and identify which filter produces an image whose feature values most closely correspond to the reference values. Figure 1 highlights the feature accuracy analysis and improvement methodology.

Overview of the methodology for improving the accuracy
of measured image features. — Overview of the methodology for improving the accuracy of measured image features. The reference signal and measured noise images are combined into a pseudo-real image typical of the target experiment. The pseudo-real image is passed through an image processing pipeline to generate a processed image. Features are extracted from the processed image and compared to features extracted from the reference signal image, enabling the evaluation of several processing pipelines to determine which improves the accuracy of the extracted features best.

Description

This experimental methodology for increasing image feature measurement accuracy consists of two major stages. The first stage is the creation of the pseudo-real images and the second stage is the optimization and evaluation of the image processing required to increase the accuracy of the measured image features in generated pseudo-real images.

In order to evaluate the accuracy of extracted image features, a known reference signal similar to ones own experiment is needed. Such a signal is acquired by imaging a representative subset of the experimental specimen with the highest image quality possible. This signal is then used to create the pseudo-real images as shown in Figure 2.

Diagram outlining the pseudo-real image creation. The reference signal (foreground) image is scaled by a factor and added to the measured background noise image to create a pseudo-real image with a specified SNR value. Note: all displayed images were auto-contrasted using the same algorithm for visual clarity.

Results

This work was motivated by a desire to understand the impact on stem cell colony classification when using image features derived from low SNR images. We devised a methodology to quantify the improvement of feature measurement for a given image pre-processing method. As a proof of concept, we chose three basic filtering techniques as pre-processing steps. We found that selecting the best filter per feature produces a 53% improvement in feature correlation with ground truth, from 0.6 to 0.92. Selecting a single filter for all features results in a 1.95% reduction in correlation and a 10% increase in residual RMSE.

To highlight the relationships between the processed image feature values and the ground truth feature values their correlation is plotted as a function of image feature, filter type, kernel size, and image SNR.

Correlation summary plot. Each image SNR block contains 4 sub-blocks, the Gaussian filter block ’Gau’, the Median filter block ’Med’, the Average filter block ’Ave’, and the No filter block ’None’. Within each filter block, the kernel size increases from bottom to top, 3 to 17. Per column within each SNR block the maximum correlation value is shown by printing the relevant kernel size.

Major Accomplishments

An understanding how to preproces the fluorescent images in order to extract accurate image features for the colony classification in the paper Bhadriraju, Kiran, et al. "Large-scale time-lapse microscopy of Oct4 expression in human embryonic stem cell colonies" Stem Cell Research 17.1 (2016): 122-129.

The paper: Michael Majurski et al., "Methodology for Increasing the Measurement Accuracy of Image Features" CVPR, 2016 (download pdf)