FKP recognition may depreciate due to the images with poor illumination, noise and pose variations. When textural feature extraction techniques are applied as it is in images, it is expected that there will be great variation in the textural features due to these extrinsic factors. Texture features can characterize regularity, randomness, directionality and coarseness properties of patterns. FKP image thus can be viewed as a texture pattern exhibiting symmetry and regularity. Hence, texture can play a predominant role in FKP recognition. LBP was originally designed by Ojala et al. (2002) for texture description, but later extended to face recognition. Recently, many more texture models such as LDN, LDTP, and LTDF_MLDN are proposed and successfully applied for pattern recognition problem.

The recognition rate mainly depends on textural features captures from the FKP image. But these textural features are severely affected by external factors. Hence, multi-resolution techniques are applied to diminish this effect. Motivated and justified by this, the combination of multi-resolution technique and texture descriptors are exploited for FKP recognition to enhance the classification accuracy.

This paper proposes a novel personal recognition system using Finger Knuckle Print. Figure 2 shows the block diagram of the proposed personal recognition system. Initially, the captured FKP images are preprocessed

by Contourlet transform which could handle FKP images with scaling, translation and rotation variant properties. This preprocessing approach also overcomes the problems that could arise due to distorted or impaired knuckle images. Secondly, the extracted ROI image of finger knuckle print is subjected to textural analyses for simultaneous extraction of textural feature information. The extracted feature vectors derived through the above methods are passed as input to the database. The texture feature of the testing images are computed and compared against the training features stored in the database using K-Nearest Neighbor and ELM classifier. Finally, the identity of a person is decided based on the matching scores derived from two different classifiers.

The rest of the paper is organized as follows: Section 2 describes the multi-resolution based preprocessing techniques in detail. Section 3 explains the various LBP based feature extraction techniques. Section 4 presents experimental results. Section 5 presents the conclusions and future enhancement of the proposed technique.

LBP is computationally simple yet very efficient local texture operator [12] . These features are invariant to monotonic gray scale changes. LBP value of a sample 3 × 3 image is deliberated by,

where g_c is the grey level value of the central pixel, g_p is the grey value of its neighbours around g_c and P is the number of neighbours. A pattern number is computed by comparing the g_c value with those of its neighbourhood. Figure 3 illustrates the basic LBP operator.

Conventional LBP requires 256 bins to store all possible patterns. The concept of uniform patterns is introduced to reduce the number of possible bins. It effectively captures the fundamental information of textures.

Adin Ramirez Rivera et al. introduced this descriptor to describe the structure and intensity transition of a texture pattern around a pixel in an image. The patterns are created with the edge response of the neighborhoods obtained by a compass mask and with the direction of the most positive and the most negative responses. The positive and negative responses are useful to determine the direction of the prominent bright and dark areas in the neighborhood. Two different compass masks such as kirsch and derivative-gaussian are experimented with LDN.

LDN assigns a six bit binary code to each pixel of an input image that represents the structure of the texture and its intensity transitions. Consequently, this pattern computes the edge response of the neighborhood using a compass mask and by taking the top directional numbers. The most positive and negative directions of those edge responses are used to describe LDN code. The more detailed explanation about LDN feature extraction is given in [13] .

Meena et al. proposed a new local texture descriptor called LDTP for face recognition under real time challenges. In LDTP, the directional features from Local Derivative Pattern (LDP) and three level descriptions from Local Texture Pattern (LTP) are fused with adaptive threshold technique (T). Figure 4 and Figure 5 represent the LDTP feature extraction in detail.

The more detailed explanation about LDTP feature extraction is given in [14] .

Local Texture Description Framework based Modified Local Directional Number Pattern is a texture descriptor that deals with the directional information of a texture pattern that is formed with pixels at dissimilar areas in an image. Computation of LTDF_MLDN code comprises four stages:

v Detection of eight sampling points around a pixel along elliptical path using LTDF,

v Calculation of positive or negative response along each direction i in a vector v,

v Computation of the directional information of prominent positive and negative responses,

v Code determination. The detailed explanation of LTDF_MLDN is explained by Reena et al. [15] .

PolyU finger knuckle print database [16] is used to evaluate the performance of the proposed method. This database is collected from 165 volunteers including 125 males and 40 females. These images are collected in two separate sessions. In each session, the subject is asked to provide six images for the left index and middle fingers, and the right index and middle fingers. In total, the database contains 7920 images with 12 images per finger. As images of a finger belong to a class, there are 660 classes. The database provides two sets of images. There are all original images in the first set, and the corresponding ROI images, with size of 110 × 220 pixels, are in another set. Some typical images are presented in Figure 6.