Phase transform for object and shape detection in digital images
View Patent ↗Object and shape detection in digital images utilizing edge detection is described. In a first edge detection approach, phase transformation is utilized in the frequency domain, such as in response to Fourier transform, followed by use of a frequency-domain phase kernel and inverse-Fourier transform. Edge detection is also provided using a phase transform in the spatial domain utilizing a convolution approach. In a second edge detection approach, phase stretching is utilized, such as in combination with phase histogramming along with thresholding and morphological operations. Numerous example images are provided illustrating benefits of the disclosed technology with different applications and under different conditions.
1. A method of performing edge detection on a digital image, comprising:
receiving a digital image within an image processing device;
applying a localization kernel prior to a nonlinear frequency dependent phase operation;
applying the nonlinear frequency dependent phase operation to the digital image in either a frequency domain or a spatial domain;
applying a phase kernel during said phase operation so that total amount of phase applied in said phase kernel determines sharpness of image edges being extracted from the digital image;
wherein edge detection is based on phase within a complex image after application of the phase kernel from said nonlinear frequency dependent phase operation; and
generating an image output as a phase image output with phase of the transformed image as output.
2. The method as recited in claim 1 , applying said nonlinear frequency dependent phase operation as a phase transformation, which is applied in either a frequency domain or a spatial domain, to the received digital image.
3. The method as recited in claim 2 , wherein said phase transformation is applied in the frequency domain comprising performing a transformation of the received digital image to the frequency domain utilizing a phase kernel, and followed by performing an inverse transform to generate a phase image in the spatial domain.
4. The method as recited in claim 2 , further comprising performing said phase transformation in the spatial domain by application of a local frequency, as a phase derivative with respect to spatial coordinates.
5. The method as recited in claim 2 , wherein said phase transformation is applied in the spatial domain in response to convolving with a spatial phase kernel.
6. The method as recited in claim 2 , wherein said phase transformation is performed utilizing a transform selected from the group of transforms consisting of Fourier Transform, Fast Fourier Transform (FFT), Discrete Fourier Transform (DFT), Discrete Sine Transform (DST) and Discrete Cosine transforms (DCT).
7. The method as recited in claim 1 , further comprising setting negative values in said generated complex phase image to zero.
8. A method of performing edge detection on a digital image, comprising:
receiving a digital image within an image processing device;
applying a localization kernel prior to a nonlinear frequency dependent phase operation;
applying the nonlinear frequency dependent phase operation as a phase transformation, applied in a frequency domain, to the digital image;
applying a phase kernel during said phase transformation in which total amount of phase applied in said phase kernel determines sharpness of image edges being extracted from the digital image;
wherein edge detection is based on phase within a complex image after application of the phase kernel from said nonlinear frequency dependent phase operation; and
generating an image output as a phase image with phase of the transformed image as output.
9. The method as recited in claim 8 , wherein said phase transformation is applied in the frequency domain comprising performing a transformation of the input image to the frequency domain utilizing a phase kernel, and followed by performing an inverse transform to generate a phase image in the spatial domain.
10. A method of performing edge detection on a digital image, comprising:
receiving a digital image within an image processing device;
applying a localization kernel prior to a nonlinear frequency dependent phase operation
applying the nonlinear frequency dependent phase operation, as a phase transformation in the spatial domain, to the digital image;
applying a phase kernel during said phase operation in response to convolving with a spatial phase kernel so that total amount of phase applied in said phase kernel determines sharpness of image edges being extracted from the digital image;
wherein edge detection is based on phase within a complex image from said nonlinear frequency dependent phase operation; and
generating an image output as a phase image with phase of the transformed image as output.