sift

arrayfire.sift(image: Array, /, n_layers: int = 3, contrast_threshold: float = 0.04, edge_threshold: float = 10.0, initial_sigma: float = 1.6, dobule_input: bool = True, intensity_scale: float = 0.00392156862745098, feature_ratio: float = 0.05) → tuple[Features, Array]

Extracts SIFT features and their descriptors from an image using the ArrayFire library.

Parameters

imageArray

The input image as a 2D ArrayFire array on which SIFT features are to be detected.

n_layersint, default=3

The number of layers per octave in the SIFT algorithm. The number of octaves is calculated based on the image size.

contrast_thresholdfloat, default=0.04

The contrast threshold used to filter out weak features in low-contrast regions of the image.

edge_thresholdfloat, default=10.0

The edge threshold used to filter out edge-like features. Higher values mean fewer features rejected based on edge-like characteristics.

initial_sigmafloat, default=1.6

The initial sigma (standard deviation) for the Gaussian blur applied to the image before feature detection.

dobule_inputbool, default=True

If True, the input image will be upscaled by a factor of 2 before processing, which helps in detecting features at larger scales.

intensity_scalefloat, default=1.0 / 255

The scale factor applied to the image intensities. Typically used to normalize the image intensities.

feature_ratiofloat, default=0.05

The maximum number of features to be detected, expressed as a ratio of the total number of image pixels.

Returns

tuple[Features, Array]

A tuple containing: - An ArrayFire Features object encapsulating the detected keypoints. - An ArrayFire Array containing the corresponding descriptors for each keypoint.

The descriptors are 128-dimensional vectors describing the local appearance around each keypoint.

Note

The SIFT algorithm is a patented technique, and its use in commercial applications may require licensing. In academic and research settings, it remains a popular choice due to its robustness and reliability.