local_stats.image

Stores the Image class, and its subclasses.

class local_stats.image.DiffractionImage(image_array: numpy.ndarray, beam_centre: Tuple[int])

Bases: local_stats.image.Image

A container for images obtained as a result of a diffraction experiment.

property pixel_chi

Returns each pixel’s azimuthal rotation for a polar coordinate mapping. This is equivalent to the typical diffraction motor chi.

property pixel_radius

Returns each pixel’s radial distance from the beam centre, in units of pixels.

class local_stats.image.Image(image_array: numpy.ndarray)

Bases: object

The base class for all images.

Attrs:

image_array:

Numpy array representing the image.

cluster(signal_length_scale: int, bkg_length_scale: int, n_sigma: float = 4, significance_mask: Optional[numpy.ndarray] = None, frac_pixels_needed: float = 0.3183098861837907) → List[local_stats.cluster.Cluster]

Returns the clustered significant pixels. Does significance calculations here under the hood.

Parameters

• signal_length_scale – The length scale over which signal is present. This is usually just a few pixels for typical magnetic diffraction data.

• bkg_length_scale – The length scale over which background level varies in a CCD image. If your CCD is perfect, you can set this to the number of pixels in a detector, but larger numbers will run more slowly. Typically something like 1/10th of the number of pixels in your detector is probably sensible.

• n_sigma – The number of standard deviations above the mean that a pixel needs to be to be considered significant.

• significance_mask – Pixels that should never be considered to be statistically significant (useful if, for example, stats are biased in this region due to a physical barrier like a beamstop).

• frac_pixels_needed – The fraction of pixels within a distance of signal_length_scale of a pixel that need to also be statistically significant for the clustering algorithm to class that pixel as being a core point in a cluster. Defaults to 1/pi.

classmethod from_file(path_to_file: str)

Instantiates an image from a path to a data file that can be opened using PIL.Image.open().

Parameters

path_to_file – The path to the image file of interest.

Returns

An instance of Image.

mask_from_clusters(clusters: List[local_stats.cluster.Cluster]) → numpy.ndarray

Generates a mask array from clusters.

Parameters

clusters – A list of the cluster objects that we’ll use to generate our mask.

Returns

A boolean numpy mask array.

subtract_background(background_array: numpy.ndarray, zero_clip=True) → None

Carried out a simple background subtraction on self.image_array. If zero_clip is true, then any pixels in image_array that are decreased below zero by the background subtraction will be clipped to zero. This is particularly useful if there’s a hot pixel in your background array.

Parameters

• background_array – A numpy array representing the background to be subtracted. OR An instance of Image representing the background.

• zero_clip – Boolean determining if the background subtracted image_array should be clipped at 0.

wavelet_denoise(signal_length_scale: int = 20, cutoff_factor: float = 0.2, max_cutoff_factor: float = 0.8, wavelet_choice: str = 'sym4') → None

Runs some wavelet denoising on the image. Without arguments, will run default denoising.

Parameters

• signal_length_scale – We would like to preferentially rotate our image away from wavelets whose length-scales are decently smaller than our signal length scale. This is the most important parameter for decimating noise wavelets. A value of 20 will kill most typical noise wavelets, but if your signal length scale is significantly larger than 20 pixels then it may be productive to increase this number.

• cutoff_factor – If any wavelet coefficient is less than cutoff_factor*(maximum wavelet coefficient), then set it to zero. The idea is that small coefficients are required to represent noise; meaningful data, as long as it is large compared to background, will require large coefficients to be constructed in the wavelet representation.

• max_cutoff_factor – The cutoff factor to be applied to signal occuring on length scales much smaller than signal_length_scale.

• wavelet_choice – Fairly arbitrary. Sym4 is the only currently supported wavelet. Look at http://wavelets.pybytes.com/ for more info. If you want a new wavelet supported, please feel free to raise an issue on the github page.