Hough-TMF 0.0.3

A package for template matching using Torch

hd-tmf

A high-dimensional template matching framework based on PyTorch.

Installation

pip install Hough-TMF

Usage

from hd_tmf import tmf

TMF Example

# generate a random template
tmp = np.random.rand(10, 20, 20)
# generate a random image
data = np.random.rand(1000, 1000)
# calculate the cross-correlation between the template and the image
corr = tmf(tmp, data, step=1, device='cpu', moves=[], batch_size=-1, save_memory=False, max_workers=4)

Parameters

• tmp (numpy.ndarray or torch.Tensor): The template to be matched.
• data (numpy.ndarray or torch.Tensor): The image to search for the template.
• step (int, optional): The step size of the convolution. Defaults to 1.
• device (str, optional): The device to perform the computation on. Defaults to 'cpu'.
• moves (list, optional): A list of moves to apply to the template before matching. Defaults to [].
• batch_size (int, optional): The batch size to use for the computation. Defaults to -1.
• save_memory (bool, optional): Whether to use half-precision floating point numbers to save memory. Defaults to False.
• max_workers (int, optional): The maximum number of worker threads to use for the computation. Defaults to 4.

Returns

• numpy.ndarray: The cross-correlation between the template and the image.

Hough Example

from hd_tmf import hough
data = np.random.randn(256, 256)
hough(data,freq=100,bandpass=[2,8],sl=[10,20],resample=1, sigma=1.3, low_threshold=3, high_threshold=6,theta=np.linspace(np.pi/2/90*10/100,np.pi/2/90*10,99), fil='bandpass', S_L=True,beta=0,kernel=(3,3))

Parameters

• data (numpy.ndarray or torch.Tensor): The image to search for the template.
• freq (int, optional): The frequency of the template. Defaults to 100.
• bandpass (list, optional): The bandpass filter to apply to the image. Defaults to [2,8].
• sl (list, optional): The size of the template. Defaults to [10,20].
• resample (int, optional): The resample rate of the image. Defaults to 1.
• sigma (float, optional): The sigma of the Gaussian filter. Defaults to 1.3.
• low_threshold (float, optional): The low threshold of the Canny edge detector. Defaults to 3.
• high_threshold (float, optional): The high threshold of the Canny edge detector. Defaults to 6.
• theta (numpy.ndarray, optional): The theta of the Hough transform. Defaults to np.linspace(np.pi/2/9010/100,np.pi/2/9010,99).
• fil (str, optional): The filter to apply to the image. Defaults to 'bandpass'.
• S_L (bool, optional): Whether to apply the Laplacian filter to the image. Defaults to True.
• beta (float, optional): The beta of the Laplacian filter. Defaults to 0.
• kernel (tuple, optional): The kernel size of the Laplacian filter. Defaults to (3,3).

License

Reference