QMT Features: September 2015
From Photons to Figures
Essential steps for accurate quantitative analysis


Between image capture and analysis lies the essential, but sometimes overlooked, step of image processing. Markus Fabich* of Olympus discusses the importance of filter methods in ensuring consistent and comparable images for quantitative digital light microscopy.

As technologies have evolved, digital image processing has become an integrated part of the microscopy workflow, often built into acquisition and analysis software. In light microscopy applications demanding clear and accurate quantification of images, the selection of optimal contrast and quantification methods is essential. Between these lies the equally essential step of image processing – the application of filter methods to raw image data. Filter methods enable the production of a final image containing fewer unwanted artefacts, which is both understandable to the human eye and suitable for quantitative analysis.

A variety of filter methods have been developed to reduce noise and enhance detail, the importance of which can sometimes be overlooked. For example, in the inspection of high grade components, where minor defects could lead to component failure, filter methods can ensure that specimens are examined consistently, accurately and efficiently, preventing defective components from continuing down the production line. In the case of research publications, non-destructive filter methods can go a long way towards ensuring image manipulation is traceable and within guidelines. Their inclusion within microscopy systems is vital to both industrial and academic microscopy workflows.

Filter methods may be classified into three categories: point, local and global operations:
Point operations - final pixel intensity is only dependent on the original intensity of that pixel. These tend to be applied across an entire image and for all images in a session, accounting for camera system and illumination bias.

Local operations - final pixel intensity is influenced by surrounding pixel intensities. They are used to apply smoothening and sharpening effects to individual images.  Global operations - final pixel intensity is dependent on all pixels in an image. Intensities are stretched or clipped to within a desired range for brightness or contrast adjustments.

The optimum method or combination of methods largely depends on the goal of imaging. For high level, detailed analysis, it is therefore imperative that the operator is able to select from a choice of filter methods for the most effective downstream image. 

Bit depth in image acquisition
Colour cannot be directly determined by camera sensors, and so RGB digital images are instead composed of greyscale intensity values in each channel – red, green and blue – stored as bits. Increasing the number of bits increases the number of intensity levels that can be stored. At least 100 intensity levels are required to ensure a smooth gradient, and to avoid the different levels becoming visible (known as posterization). An 8 bit image, with 256 levels of intensity, is more than sufficient for this.

However, the application of filters can lead to the loss of raw image data, and so microscope cameras are designed to capture 12 bit or 16 bit depth images to avoid visible posterization. In order to correctly display these higher bit depth images for accurate observation and analysis, they need to be temporarily remapped to the 8 bit depth of a monitor. To avoid clipping, where the intensity levels above 255 are displayed as pure white, alternative mapping ratios are applied, such as 16:1 for 12 bit images. Here each consecutive set of 16 intensities is displayed as a single level, allowing the raw image to be displayed without visible clipping.

Microscope operators can also take advantage of mapping by using intensity scaling to manipulate an image, deliberately condensing the range of intensities to increase contrast or shifting all intensities to increase or decrease brightness in the raw image. This type of point operation is applied across the whole image in a linear fashion or on a curve.

Background and shading correction

Background and shading corrections are more permanent changes than intensity scaling, applied to correct for background generated internally within the imaging equipment, or inconsistent lighting. Background correction is a point operation that allows the camera dark image (image data captured without intentional illumination) to be subtracted. An operator can manually apply this using the image histogram, selecting a threshold greyscale value that allows separation of the background and actual signal peaks. Similarly, variations in shading due to inconsistent illumination can be corrected automatically using a blank image to create a flat line profile of illumination (Figure 1).

Noise reduction

Noise reduction methods are local operations, where a pixel is influenced by its neighbours. From grids of values (convolution kernels) a mathematical operation is applied to a central target pixel based on these intensities. Noise reduction filters smoothen the transition in intensity from pixel to pixel and mean and median filters, for example, apply the mean or median value from the pixels contained in the convolution kernel. The resulting effect is a reduction in noise, but also a reduction in sharpness of edges by smoothening of contrast.

For identifying and removing random high frequency noise internal to the camera system, the Gaussian filter can be a powerful tool. It relies on the smallest detectable particle on a sample to cover at least three adjacent pixels, creating a point spread function. Random noise appears as single high contrast pixels (Figure 2), and therefore can be differentiated from real signals that are diffraction limited. This makes the Gaussian filter superior to the Mean or Median filters in eliminating noise while retaining point spread functions, especially when imaging highly detailed samples that require careful quantitative analysis, such as high grade electronics.

On the edge - accurate quantitative analysis
Smoothening allows an operator to analyse an image with less bias from random noise. This process, however, also affects high contrast edges, which become less distinct. Edge detection filters are multi-stage local operations based on convolution kernels enhancing the edges of objects. Enhanced contrast ensures quantitative measurements remain as accurate as possible.

Following the application of the smoothening filter, the signal is converted from low-pass to high-pass and merged with the original. As a result, dark edges become darker and the light edges become lighter, increasing contrast for pixels not already at the lower or upper intensity limit.

The convolution kernels are designed such that adjacent pixels of equal intensity are not affected, and therefore selective to edges. Random high-frequency noise, however, can affect the outcome and it is therefore generally recommended that noise removal is performed beforehand.

Sobel and Laplace are two complex, but powerful edge detection filters that can significantly alter the image (Figure 3). Although use of these filters generates images that do not reflect what is being observed through the eyepieces or on the monitor, they do accurately represent edges, as required for quantitative analysis.

Integrating filter methods into microscopy workflow

Filter methods have become an essential part of quantitative digital microscopy. In response to this trend, filter methods are integrated directly into the latest digital light microscopes, such as the Olympus DSX series for non-destructive industrial inspection and the LEXT OLS4100 metrology microscope. This is also the case for specialised materials science software, where the Olympus Stream software family provides a full set of image processing methods at the user’s fingertips.
This guides the operator through acquisition, processing and analysis, reducing hands-on time and making best use of the high-quality optics. Following acquisition, options enable background and noise reduction, as well as edge enhancing to produce images for quantitative analysis. Previews are provided to quickly test the comprehensive range of filter methods before alterations are applied non-destructively, ensuring the original raw image data is retained. Including these functions in microscopy systems has allowed quick and easy integration of filter methods into the workflow of operators with all levels of experience.

Summary
For industrial inspection microscopy, the application of the most appropriate filter method can vastly improve the ease and accuracy of downstream quantitative analysis. These methods may include point, local or global operations, which can be used to reduce random noise and enhance detail both locally and globally across an image.
The availability of modern image processing software has greatly facilitated the accessibility of filter methods, and their integration into microscopy systems has allowed microscope operators to quickly and easily choose and apply the correct filter method from a comprehensive selection. All without risking permanent alteration of the original raw image data. Integrated systems ensure consistency across a range of images from multiple imaging sessions, and their comprehensive software has come to play an essential role in modern light microscopy.
www.olympus-europa.com
  
You can now view all QMT Magazine issues on your favourite tablet or smart phone.
Download the free Quality Manufacturing Today App from the Apple iTunes App Store or from QMT Magazine on Google Play.

Rob Tremain Photographer
www.4exposure.co.uk
slideShow
Click above to see full page display and links to QMT articles.
Untitled
Mitutoyo logo
Bowers logo
Prodim logoi
TCT Inspex 2016 logo
Aberlink logo
Control logo
Creaform logo
Vision 2016 ad