Finger extraction, finger image automatic registration, and finger identification by image phase matching

https://doi.org/10.1016/j.amc.2006.10.074Get rights and content

Abstract

In this research, a new technique is used to extract the thumb, index, middle, ring, and small fingers and to perform a person’s identification. To allow the finger geometry to be more salient when performing the finger image comparison, the fingers are scaled to various sizes. For reducing the number of finger-image-files in the system, a person’s entire finger-images are placed in one file. The hand is fixed each time when a picture is taken and one can assume that each time when the hand image is taken, the acquired finger images are the same as the previously acquired ones. Since the pictures are the same, after the fingers are extracted from the hand image, one can use the acquired fingers to identify different people. In this research, the developed algorithm of the auto-registration technique can find the precise location of the finger image – including the centroid of the finger image and the orientation of the finger image. The finding of the position and the orientation of the finger image are conducted automatically and without any further human effort. After finding the positions of the finger images, image rotating, image shifting, and image interpolating techniques are used to align different finger images to the same position and the same orientation for comparison. The extracted finger image contains many useful geometrical features. One can use these features to do finger image identification. Since the entire finger images are aligned to the same position and the same orientation, the image phase-matching technique is used to examine the difference between two finger images. The image phase-matching technique involves complex number manipulation and also finds the most salient feature of the resultant images.

Introduction

During the past 20 years, researchers invested a lot of effort to develop different techniques to identify hand images. This past work includes hand geometry [1], [2], [4], [5], [12], [15], [16], [17], [18], middle finger crease pattern matching [6], various finger size measurements [7], [14], various finger lateral view size measurements [14], vein pattern [9], eigenpalm [10], implicit polynomials [13], algebraic invariants [13], Karen invariant computation [13], line interception and slope comparisons [19], control point selection [8], [19], coarse to fine strategy [3], B-Spline [11], watershed transform [9], HMM [16]. However, some are very sensitive to the noise [6], [19]; some have very complicated mathematical models [13] and some have very complicated neural training algorithms.

In the previous research, when performing the hand geometry matching [1], [2], [4], [5], [12], [15], [16], [17], [18], in order to recognize the hand image, every time the hand needs to be placed in a precise certain fixed position – thus the camera can capture the same hand image. In this research the hand need not placed in a certain precise fixed position. The wavelet technique is used to find the finger-to-finger valley be the past research [3], [4]. In this research, the developed recognition algorithm automatically calculates and checks the finger-edge energy response signals and selected the high energy response signals to find the finger-to-finger valleys of the hand image automatically. The algorithm developed in this research finds the finger-to-finger valleys more accurately and more efficiently. This research amputates the finger from the hand image automatically and calculates the feature and shape of each individual finger image separately. This research uses the amputated finger to generate more original finger features, which contain more opulent data than the finger crease pattern [6] and rude finger shape matching methods [3], [4], [14]. Implicit polynomials are very difficult to describe the finger shape by the power of seven polynomial functions. In the previous research, the palm-print [3], [4], iris, fingerprint, face and vein pattern [9] are also used to identify different persons.

In this research, the new technique is used to extract the thumb, index, middle, ring, and small fingers and to perform the person’s identification. To allow the finger geometry more salient when performing the finger image comparison, the fingers are scaled to various sizes. For reducing the number of finger-image-files in the system, one person’s entire finger-images are placed in one file. The hands 8 are fixed each time when a picture is taken and one can assume that each time when the hand image is taken, the acquired finger images are the same as the previous acquired ones. Since the pictures are the same, after the fingers are extracted from the hand image, one can use the acquired fingers to identify different people.

In this research, the image automatic registration algorithm finds the orientations and positions of the extracted finger images automatically. In order to perform the phase-matching comparison of two finger images, the centroids of the finger images need to be found. The images used in this research are the 128 by 128 images. One needs to perform the image movement to move the image to the center of the 128 by 128 image frame – i.e. we need to shift the image to allow the centroid of each image to be shifted to the location (64, 64) in the 128 by 128 image frame. Furthermore, the major axes of the finger images need to be found and we need to move the image to allow the major axes of the finger images to be aligned to a straight position. Since every image is shifted to the center of the 128 by 128 image frame and the major axis of each image is aligned to the same vertical position, one can perform the image phase-matching comparison on each image. The above steps are performed by the computer itself and no further human involvement is required.

After the finger images are shifted and rotated to the same position, one can use a phase-matching discriminator to judge whether or not the finger images are the same finger images. The phase-matching technique involves complex number manipulation. The techniques – database SQL searching and database manipulating, image dilating, and image interpolating, are used to perform finger image recognition.

This paper consists of five sections. Section 2 extracts in finger images. Section 3 describes the technique of the Image auto-registration. Section 4 describes image phase matching. Section 5 concludes this paper.

Section snippets

Extraction of finger images

Fig. 2.1 shows the shelves for taking hand-images. The illuminations are from the left lateral and the bottom of the shelves. By adjusting the lights one can control the illuminations of the hand-image. In the middle shelf, several pegs are used to peg the person’s hand to a certain position. This will make the hand inertial when the hand-image is taken. Fig. 2.2 shows the hand images. Fig. 2.3 shows the extracted hand edge images. In Fig. 2.4 kx represents the pixel of the

Image auto-registration for image comparison

In order to perform the comparison of two images, the centroids of the finger images need to be found. The images used in this research are the 128 by 128 images. One needs to perform image movement to move the image to the center of the 128 by 128 image frame – i.e. we need to shift the image to allow the centroid of each image to be shifted to the location (64, 64) in the 128 by 128 image frame. Furthermore, the major axes of the finger images need to be found and we need to move the image to

Image phase matching

Eq. (4.1) shows the function which transfers the function f(x, y) to function F(u, v) and Eq. (4.2) shows the function which transfers the function g(x, y) to function G(u, v). Fig. 4.1 shows the various f(x, y) and its corresponding F(u, v) images. Eq. (4.3) shows the relationship of image phase matching between – F(u, v), conjugated G(u, v), f(x, y), and g(x, y). Symbol ∘ represents the image phase matching. Fig. 4.2 shows the phase-matching images of the two images f(x, y) and g(x, y). By checking Fig.

Results and conclusions

In this research, totally three hand images are taken for each person and 40 persons are included in the experimental test. The hand images used in this system are 120. Regarding these 120 hand images as a whole and running these 120 hand images in one batch – will take a very long time to get the job done. Due to a software glitch, the running process might be halted before the job is completed. Consequently, one might endlessly start the procedure again and again and every time from square

References (19)

There are more references available in the full text version of this article.

Cited by (4)

  • A lattice Boltzmann method applied to the fluid image registration

    2019, Applied Mathematics and Computation
    Citation Excerpt :

    Indeed, the need of registration tools is inevitable in many applications, such as: remote sensing, astronomy, chemistry and also medical imaging. Especially, image registration technique is recently used in disease evaluation (tumors), when comparing an image to its counterpart in a given database [4–7]. For a detailed overview about potential image registration applications, see e.g. [8].

  • Accurate point matching based on multi-objective Genetic Algorithm for multi-sensor satellite imagery

    2014, Applied Mathematics and Computation
    Citation Excerpt :

    Image registration is the process of geometrically aligning two images in a way that corresponding pixels in the two images correspond to the same physical region of the scene being imaged [1–4].

  • Efficient transportation system using internet of things

    2015, International Journal of Applied Engineering Research

National Science Council, Taiwan, supported this work under grant NSC 95-2221-E-212-003.

View full text