Invariant neural-network based face detection with orthogonal Fourier-Mellin moments

Orthogonal Fourier-Mellin (OFM) moments have better feature representation capabilities, and are more robust to image noise than the conventional Zernike moments and pseudo-Zernike moments. However, OFM moments have not been extensively used as feature descriptors since they do not possess scale invariance. This paper discusses the drawbacks of the existing methods of extracting OFM moments, and proposes an improved OFM moments. A part of the theory, which proves the improved OFM moments possesses invariance of rotation and scale, is given. The performance of the improved OFM moments is experimentally examined using trademark images, and the invariance of the improved OFM moments is shown to have been greatly improved over the current methods.

show abstract

“…3 image the moments differ. Figure 1 shows the process of extracting the OFM moments from an image [16]. First, the input image is binarized.…”

Section: Invariance Analysis Of Orthogonal Fourier Mellin Momentsmentioning

confidence: 99%

“…3,16 They have been proven to be superior to other moment functions such as Zernike moments and pseudo-Zernike moments in terms of their feature representation capabilities and robustness in the presence of image quantization error and noise. 14 OFM moments offer more feature vectors than Zernike moments and pseudo-Zernike moments.…”

Section: Introductionmentioning

confidence: 99%

Improvement and Invariance Analysis of Orthogonal Fourier–Mellin Moments

Yao

Peng

Ren

et al. 2003

Int. J. Patt. Recogn. Artif. Intell.

View full text Add to dashboard Cite

show abstract

“…These moments are rotationally invariants and have the ability to describe spatial frequency components of an image. Due to their elegant characteristics, OFMMs are used in image processing and pattern recognition applications [3][4][5][6][7].…”

Section: Introductionmentioning

confidence: 99%

Fast computation of orthogonal Fourier–Mellin moments in polar coordinates

Hosny

Shouman

Salam

2009

J Real-Time Image Proc

View full text Add to dashboard Cite

Fast, accurate and memory-efficient method is proposed for computing orthogonal Fourier-Mellin moments. Since the basis polynomials are continuous orthogonal polynomials defined in polar coordinates over a unit disk, the proposed method is applied to polar coordinates where the unit disk is divided into a number of non-overlapping circular rings that are divided into circular sectors of the same area. Each sector is represented by one point in its center. The implementation of this method completely removes both approximation and geometrical errors produced by the conventional methods. Based on the symmetry property, a fast and memory-efficient algorithm is proposed to accelerate the moment's computations. A comparison to conventional methods is performed. Numerical experiments are performed to ensure the efficiency of the proposed method.

show abstract