Process Big Images Efficiently Using Partial Images or Lower Resolutions

This example shows how to process a big image quickly using two strategies that enable computatations on smaller representative samples of the high-resolution image.

Processing big images can be time consuming. This makes iterative development of algorithms prohibitively expensive. There are two common ways to shorten the feedback cycle: iterate on a lower resolution image or iterate on a partial region of the big image. This example demonstrates both of these approaches for creating a segmentation mask for a big image.

If you have Parallel Computing Toolbox™ installed, then you can further accelerate the processing by using multiple workers.

Create a bigimage. This examples uses a modified version of image "tumor_091.tif" from the CAMELYON16 data set. The original image is a training image of a lymph node containing tumor tissue. The original image has eight resolution levels, and the finest level has resolution 53760-by-61440. The modified image has only three coarse resolution levels. The spatial referencing of the modified image has been adjusted to enforce a consistent aspect ratio and to register features at each level.

bim = bigimage('tumor_091R.tif');

Display the big image by using the bigimageshow function.

bigimageshow(bim);

Accelerate Processing Using Lower Resolution Image

Many big images contain multiple resolution levels, including coarse lower resolution versions of the finest high-resolution image. In general, the distribution of individual pixel values should be roughly equal across all the levels. Leveraging this assumption, you can compute global statistics at a coarse level and then use the statistics to process the finer levels.

Extract the image at the coarsest level, then convert the image to grayscale.

imLowRes = getFullLevel(bim,bim.CoarsestResolutionLevel);
imLowResGray = rgb2gray(imLowRes);

Threshold the image into two classes and display the result.

thresh = graythresh(imLowResGray);
imLowResQuant = imbinarize(imLowResGray,thresh);
imshow(imLowResQuant)

Validate on the largest image. Negate the result to obtain a mask for the stained region.

bq = apply(bim,bim.FinestResolutionLevel, ...
    @(im)~imbinarize(rgb2gray(im),thresh));

Visualize the result at the finest level.

bigimageshow(bq,'CDataMapping','scaled');

Accelerate Processing Using Using Partial Regions of Big Image

Another approach while working with large images is to extract a smaller region with features of interest. You can compute statistics from the ROI and then use the statistics to process the entire high-resolution image.

% Zoom in on a region of interest.
bigimageshow(bim);
xlim([2400,3300])
ylim([900 1700])

Extract the region being shown from the finest level.

xrange = xlim;
yrange = ylim;
imRegion = bim.getRegion(1,[xrange(1),yrange(1)],[xrange(2),yrange(2)]);
imshow(imRegion);

Prototype with this region then display the results.

imRegionGray = rgb2gray(imRegion);
thresh = graythresh(imRegionGray);
imLowResQuant = ~imbinarize(imRegionGray,thresh);

imshow(imLowResQuant)

Validate on the full big image and display the results.

bq = apply(bim, bim.FinestResolutionLevel,...
    @(im)~imbinarize(rgb2gray(im), thresh));

bigimageshow(bq,'CDataMapping','scaled');

Accelerate Processing Using Parallel Computing Toolbox

If you have the Parallel Computing Toolbox™ installed, then you can distribute the processing across multiple workers to accelerate the processing. To try processing the image in parallel, set the runInParallel variable to true.

runInParallel = false;
if runInParallel
    % Open a pool
    p = gcp;
    % Ensure workers are on the same folder as the file to be able to
    % access it using just the relative path
    sourceDir = fileparts(which('tumor_091R.tif'));
    spmd
        cd(sourceDir)
    end
    % Run in parallel
    bq = apply(bim,bim.FinestResolutionLevel, ...
        @(im)~imbinarize(rgb2gray(im),thresh),'UseParallel',true);
end

See Also

| |