Pavankumar Prathipati (Author)Dept. of Information Technology SRH Hochschule HeidelbergHeidelberg, [email protected] Dr Ing Christof Jonietz (Author)Dept. of Information TechnologySRH Hochschule HeidelbergHeidelberg, [email protected] Abstract— For the past decades, there has been a tremendous growth in thestudy of segmenting the skin regions in color images. Hand Recognition Via skincolor segmentation is one of the challenging issues in image processing.
Inthis paper, we make a study in skin color segmentation to recognize the Handand can be used to extract a finger print with palm, if possible with differenthand gestures and perform various operations on it like recognition of finger,box bounding fingers, palm with out fingers, the lines appearing on bondingbox.Keywords-component; skin tone segmentation I. Introduction The appliance of skin segmentation isbeing used commonly in all areas of computer science and its purposes.
Detecting and monitoring face and palms are major for gesture consciousness andhuman computer interaction. The purpose of this skin segmentation is toidentify contact less palm and finger print Verification, Identificationprocess is considered here. This is contact less figure capture will enhancethe security, minimize spoofing and evasion, make the control less complicated forpassengers. Figure (1) Inhuman-computer interaction (HCI) recognition of hand gestures is one of theimportant parts and it grabs the attention of many researchers.
In order torecognize the hand gestures there are some devices, those devices use aninfrared projector and camera and a special algorithm to track the moments ofobjects in three dimensions, but there is no robust solution by using colorcamera only.In order to recognize the hand gestures,the first task is to identify the where the hand is located in the image. Butin some methods to identify the hand they use skin tone color without using anytextural features. These assumptions have negative impact on detection quality,so in order to improve the detection quality up to 90% we use more complexapproaches. So, here contactlessfingerprint capturing, finger detection and finger extraction are consideredhereThere are several approaches for fingerdetection, finger detection is based top-hat transform, which usesmorphological operators on a hand blob by subtracting the results from theoriginal hand blob. However, a reliable segmentation of the hand palm orfingers in the RGB image is necessity of many algorithms that analyze a binaryhand blob image and In given color image to extract the skin related pixelsdepends on the correctness of the algorithm.
The workflow comprises palm detection inorder to detect the skin region with in given image to the detection of fingertips. In order to detect the palm detection and finger detection there is amachine learning based algorithm and for fingertip detection a fast-geometricshape based approach, has been designed to reconstruct long lines alongfingers. Commonly image processing simulation isdone by MATLAB through converting image in to 2D or 3D matrices to performdifferent operations on it. In order to develop a real-time image processing application,we use Qt-creator this qt will program a finger print skin tone segmentation.Here the skin segmentation is a method of segmenting the pixels of a photographin to skin and non-skin regions. This algorithm must be able to make a decisioncouple of particular pixel in to skin regions or non-skin areas.
However, inour project we use Built in camera or web camera II. Workingmethodology First, we must define how our projectwill work on the basis of two algorithms. One algorithm is skin color basedhand segmentation and other algorithm is finger print extraction from segmentedpalm or finger.
A. Skin color based hand segementation Identify applicable sponsor/s here. If no sponsors, delete this text box.
(sponsors) In order to the detect the palm the stepis to determine the subset of pixels which belong to the hand. Thissegmentation step allows for sub sequential analysis of hand and to search forfinger tips.One approach for segmenting the bodyparts is using skin color. Several approaches are there to detect pixels ofhuman skin without prior knowledge on images content. In our case skin colorestimation and segmentation can be done in a more robust way. 1) Pixel based skin tone detection. For pixel-based skin tone segmentation we usethe proposed algorithm by********. There are alternative approaches which tryto remove the illumination component from images to obtain an illuminationinvariant color classifier and that illumination is also important feature forpixel wise skin tone classification.
The algorithmfirst determines a gray scale map of RGB color image and given a usualtransformation matrix as By standardproduct operation The resulting is1Dimage and Idash is considered as the gray scale map of the original imagetaking in to account of all color channels. Additionaly to Idash a secondillumination map is determined considering only blue and channel componentsonly. Red channel is eliminated from grayscale map, since its most contributingone in skin pixels Now skin colorprobability map is generated by pixel wise signal error calculation as Finally, in agiven largest training dataset containing skin regions from persons of a rangewith extreme variation of lighting effect, the interval of E(x) for themajority of skin pixels has been estimated. Given a lower,upper bound, skin pixels are classified as follows The important reason for choosing theskin detection method is its efficiency and real-time capability and advantageis no complex color space is needed and for very low computational load thereis reduction of color space dimensionality from 3D(RGB) to 1D (normalizedgrayscale) and has efficient pixel wise classification.(b)image after skin color segmentationFigure2(a) figure2(b) Analysisfor clutter Removal/De-noising. in the given segmenting image of skinpixel as shown in fig(b), a connected component analysis is performed for blobanalysis and clutter removal.
All small segments in an small area related topalm region of interest(ROI) are classified as cluster. In our examples wechoose 5% of ROI area classified as cluster. All segments outside of the palm detectorROI are eliminated in further processing and finally containing the blobs ofskin colored areas of significant size only. Post-Processing/ Morphology.In a given candidate hand segment asdescribed above, in this post-processing morphological filters are applied forhole filling. So, there by a background flood will approach, followed by abinary image inversion is applied and the result is shown below in Fig © Figure2 (c) (c) clutter removal, blob filling andmorphology 2) Fingertip extraction The main concept of finger detectionalgorithm is to involve left and right edges to fingers (edge pairing) and toextract by their respective tips and angles.The edge pairing algorithm is dividedinto the following stepsHandpalm segmentation.
The input RGB image(a) issegmented in order to extract the finger edges Figure3(a)Hand palm segmentation is based on thealgorithm described above in pixel based skin tone detection.Convertingthe binary blob image into a contour line.In order to process and examine the blobscontained in the binary input image as shown in figure (b) Figure3(b) And from the above image, the blobs areconverted into contour lines as shown in figure (c) Figure 3(c)A contour is a sequence of a pixellocated along the boundaries of the blob. Line segmentation of the contour line.The most important feature of fingerrecognition is to recognize the long lines along them. In order to detect suchlines and decrease the presence of fingers and it is necessary to suppuratethem from the contour line of a blob by suppurating the line segments with in acontour. By analyzing the variation of the tangent angle on the contour we candetect the line segments.
The variation is low with in the line segment and thetangent angle remains almost constant.Formingfingers by matching edges.The formation of fingers based onpreviously computed lines segments and line segments provided by edge pairing.A pair of edges is consequently explained the left and right edges of thefingers. The method is explained in the below figure (d) Figure3 (d)From the above figure the line segmentsused for reconstruction of paired edges are plotted in the same color. In thismethod an edge can only combined with another edge and I f the edges are notpaired then it is discarded and are not recognized as a finger.In order to find the best match for eachedge, match quality metric for a hypothetical pair of edges is calculated, whichdepends on the orientation of edges, distance between center points, themaximum allowed pairing angle and the length of edges. III.
obtained results results are provided for the pixel basedskin tone detection and the finger extraction algorithm based on assigningassociated edges through pairing in the following the trained detector localizes the palmsindependent of subjects and shape. Robustness against closed and spread fingersis demonstrated. results of theproposed finger extraction algorithm based on assigning associated edgesthrough pairing are presented. In the below figure (xx) the finger axes and ROIare shown. The ROIscontaining that part of fingers used for a later biometric verification oridentification process. After calculating the fingertip and finger axis thesize of the ROI is determined dynamically to take in to account individualfinger sizes. The width of theROI depends or based on the width of the finger and is determined by examiningthe binary blob image on a virtual line perpendicular to the finger axis.
Theheight of ROI depends on its width and is determined by multiplying the fingerwidth with a constant factor. Here differentresults for palms and fingers are presented. The mostimportant recognition feature of fingers is to recognize the long lines alongthem. Since the algorithm has designed to reconstruct the long lines along thefinger and only for separate fingers can be reconstructed such that resultswith spread fingers are shown here. Since extraction of edges is based onsegmented palm in order to deduce the presence of fingers and the segmentationof palm is one of the important feature here.
If these two conditions arefulfilled, the dynamic ROI can be determined reliably. For single finger: Figure4(a) For two fingers: Figure 4(b) For threefingers: Figure 4(c) For four fingers: Figure 4(d) 1V.Conclusion:In this paper, contactless palm andfinger detection for the biometric recognition process has been considered.Evaluation based on a measurement campaign in different indoor and outdoorscenarios demonstrate the suitability of the geometric approach by edge pairingin extracting the finger tips for a biometric identification and verificationprocess. Softwaresused: In this project code is implemented byusing qt server which is better choice for implementing real time computerhuman interaction based applications. Open CV helps to add more flexibility toour source which will help us to modify the code to our needs. Advantages: The biggest important reason for choosingskin-tone detection method is efficiency and Real-time capability and reductionof color space dimensions from 3Dimensions to 2Dimensions allows for lowcomputational load and efficiency pixel wise. Moreover, we don’t use expensive cameranormal high-resolution HD camera is used to implement the algorithms.
References: 1) X. Bai, L. J. Latecki, and W.-Y.Liu. Skeleton Pruning ContourPartitioning with Dis-creteCurve Evolution.
Pattern Analysis and Machine Intelligence, IEEE Transactionson, 29(3):449–462, March 2007. 2) A. Cheddad, J. Condell, K.Curran, and P.McKevitt. “A new color space for skin tone detection”. IEEEInternational conference on image processing (ICIP), 2009 16th,pages 497-500, Nov 2009.
3) S.B. Jeema, M.
Hammami, and H.Ben-Abdallah. Data-mining process: application for hand detection in contactfree settings.
Image processing, 7(8):742-750, Nov 2013. 4)