Image Processing

Collection of image processing algorithms and techniques that could be used and widely applied to many of our datasets.

Functions:

bin_2d(im, bin_size[, stride, pad_value, debug])	Bins a 2d array by a square kernel length (bin_size*2 + 1).
fill_and_label_rois(curr_nuclei)	Fills and labels ROI outlines using unique ints for each region.
four_color_theorem(mask)	Converts a n-labeled image to an up to 4 color image.
four_color_to_unique(mask[, debug])	Converts an n-label image into a
kmeans_threshold(im, k, n_brightest_clusters)	Given an image, this function will apply k-means clustering and return the mask that includes the n brightest clusters :param im: intensity image.
normalize(im)	Normalizes an image to 0 and 1 by subtrating min value and dividing by new max value.
remove_horizontal_vertical_edges(im[, ...])	This function removes horizontal or vertical edges in the image.
remove_small_areas_fill_regions(mask[, ...])	Function was created to clean up masks that may have stray pixels or regions.
rgb2gray(mask_rgb[, debug])	# RGB to Grayscale ## Code logic: - Essentially, read in mask only image (not composite) and figure out unique tuples that represent singular RGB values.
rgb2labels(im[, debug])	Converts a 3ch rgb mask image into a 1ch uint16 mask image.
sum_pool_3d(im, bin_size[, stride, ...])	Sums the pixel intensity by a given kernel, reducing output image dimensions.

cell_analysis_tools.image_processing.bin_2d(im, bin_size, stride=1, pad_value=0, debug=False)

Bins a 2d array by a square kernel length (bin_size*2 + 1).

Parameters:

• im (ndarray) – 3d ndarray containing FLIM data (x,y,t).

• bin_size (int) – number of pixels to look before and after center pixel. kernel = bin_size + 1 + bin_size

• stride (int, optional) – Number of pixels to move when raster scanning the image. The default is 1.

• pad_value (int, optional) – value to pad the image with. The default is 0.

• debug (bool, optional) – Show debugging output. The default is False.

Returns:

im_binned – binned 3d array of decays

Return type:

ndarray

cell_analysis_tools.image_processing.fill_and_label_rois(curr_nuclei)

Fills and labels ROI outlines using unique ints for each region.

Parameters:

curr_nucelei (param) – Current image to process

Returns:

output – ROIs filled and labeled with unique int representations

Return type:

ndarray

cell_analysis_tools.image_processing.four_color_theorem(mask: ndarray) → ndarray

Converts a n-labeled image to an up to 4 color image.

Parameters:

mask (np.ndarray) – labeled mask with unique rois.

Returns:

• four_color_mask (np.ndarray) – returns a labeled image that uses up to 4 colors.

• solution_nodes (dict) – dictionary of {value : color} for each unique roi value

cell_analysis_tools.image_processing.four_color_to_unique(mask: ndarray, debug: bool = False) → ndarray

Converts an n-label image into a

Parameters:

• mask (np.ndarray) – labeled mask.

• debug (bool, optional) – show output re-labeled mask. The default is False.

Returns:

output_mask – Mask relabeled to have unique roi values for each connected component

Return type:

np.ndarray

cell_analysis_tools.image_processing.kmeans_threshold(im, k, n_brightest_clusters, show_image=False)

Given an image, this function will apply k-means clustering and return the mask that includes the n brightest clusters :param im: intensity image. :type im: array-like :param k: number of clusters for k means algorithm :type k: int, :param n_brightest_clusters: number of brightest clusers to keep. Must be < k. :type n_brightest_clusters: int, :param show_image: When debugging,this will display the original image next to the k means thresholded image. The default is False. :type show_image: bool, optional

Returns:

mask – binary mask

Return type:

int

cell_analysis_tools.image_processing.normalize(im)

Normalizes an image to 0 and 1 by subtrating min value and dividing by new max value.

Parameters:

im (array-like) – intensity image.

Returns:

array normalized to 0 and 1.

Return type:

array-like

cell_analysis_tools.image_processing.remove_horizontal_vertical_edges(im, disk_size=20, debug=False)

This function removes horizontal or vertical edges in the image. Designed for removing grid patterns in PDMS scaffolds during time lapse imaging.

Parameters:

• im (ndarray) – intensit image with grid pattern

• disk_size (int) – size of filter used to filter low frequency components of 2d image

• debug (bool, optional) – output intermediate images , by default False

Returns:

filtered image with grid pattern removed

Return type:

np.ndarray

cell_analysis_tools.image_processing.remove_small_areas_fill_regions(mask: array, region_min_size: int = 100, footprint_area_closing: int = 2, debug=False)

Function was created to clean up masks that may have stray pixels or regions.

Parameters:

• mask (np.array) – original mask.

• region_min_size (int, optional) – minimum size of connected component to be removed. The default is 100.

• footprint_area_closing (int, optional) – When merging images radius of disk to use. The default is 2.

• debug (TYPE, optional) – Enable/Disable display of intermediate images for debugging function. The default is False.

Return type:

np.array of revised mask

cell_analysis_tools.image_processing.rgb2gray(mask_rgb, debug=False)

# RGB to Grayscale ## Code logic: - Essentially, read in mask only image (not composite) and figure out unique tuples that represent singular RGB values. - A typical RBG value to produce a color value can be broken up into (R, G, B) insity values. - Tuples are preferable over numpy arrays as they can be easily condensed into unique elemnts using a set and then converted back into a list (they are hashable). - After uniques are found, build a dictionary with values from 1 to number of unique RGB values in the images. These will become the grayscale int values that we then plot. The resulting array should automatically upscale from 8-bit integer if needed (>256 unique elements). - This only works for mask images, NOT overlays!

Parameters:

• mask_rgb (np.ndarray) – rgb image.

• debug (bool, optional) – Enable debugging output. The default is False.

Returns:

result – Grayscale image.

Return type:

np.ndarray

cell_analysis_tools.image_processing.rgb2labels(im, debug=False)

Converts a 3ch rgb mask image into a 1ch uint16 mask image.

Parameters:

• im (ndarray) – input 3ch RGB image of rois.

• debug (bool, optional) – Enable display of intermediates. The default is False.

Returns:

mask – labels mask from RGB image.

Return type:

ndarray

cell_analysis_tools.image_processing.sum_pool_3d(im, bin_size, stride=1, pad_value=0, debug=False)

Sums the pixel intensity by a given kernel, reducing output image dimensions.

Parameters:

• im (ndarray) – 3d array containig FLIM data (x,y,t).

• bin_size (int) – number of pixels to select before and after, bin of 3 is a 7x7 kernel

• stride (int, optional) – number of pixels to advance when doing raster calculations. The default is 1.

• pad_value (int, optional) – Value to pad the edges with if kernel needs it. The default is 0.

• debug (bool , optional) – Show debugging output. The default is False.

Returns:

im_sum_pool – 3d ndarray that holds a summed version of the image resized image.

Return type:

ndarray