선택한 단어 수는 입니다.
최소 단어 이상 선택하여야 합니다.
최소 단어 이상 선택하여야 합니다.
선택한 단어 수는 30입니다.
최대 10 단어까지만 선택 가능합니다.
최대 10 단어까지만 선택 가능합니다.
문제 정의
- This paper measures and evaluates various features from passport images used in an experiment. And, for the improvement of processing speed and reliability, this paper applied Luminance, Edge value and RGB color distribution to the correlation measurement for photo verification and, Hue, I value of TIQ color model, Cb-value of YCbCr color model for the face verification.
- This paper proposes face verification of the passport image for precise and reliable discrimination of passport forgery[11]. The combination of face verification and photo verification is to enable discrimination in forged passports by image synthesis of facial areas in the photo.
가설 설정
- Step 1. Define xk as the fc-th input vector, wz as the connection weight vector of z-th cluster in the neural network. Step 2.
- Define xk as the fc-th input vector, wz as the connection weight vector of z-th cluster in the neural network. Step 2. Perfbnn the Exclusive-NOR operation on the newly-given xk and 叫, and select a winner node with maximum value.
제안 방법
- Analyzing the results shown in Table 2 and 3, for the improvement of processing speed and reliability, this paper applied selective features to the correlation measurement which is as follows: Luminance, Edge and RGB for photo verification, and Hue, YIQ-I, YCbCr-Cb for face verification.
- and Cr-value of YCbCr color distribution. And in this paper, the features were declared as Luminance, Edge, Hue, RGB, RGB-N, YIQ_I, YIQ_Q, YCbCr-Cb and YCbCr-Cr, respectively. The experiment for photo and face verification measured and compared the correlation between the 16 passport photos saved in the database and the 112 photos extracted from forged and varied passport images.
- The proposed system extracts and deskewes the code areas from passport images by using properties of passport code areas. And this paper proposed the novel ART algorithm and applied it to the extracted code areas for code recognition. After compensating heuristically the recognition result, the discrimination of forged passports is achieved by using the photo and face verifications between the photo extracted from the passport image and the picture retrieved from the database based on the recognized codes.
- images used in an experiment. And, for the improvement of processing speed and reliability, this paper applied Luminance, Edge value and RGB color distribution to the correlation measurement for photo verification and, Hue, I value of TIQ color model, Cb-value of YCbCr color model for the face verification. The overall process of code recognition and face verification for the passport image is shown in Fig.
- Comparing the candidate areas with the regular size of the skin-color area and the face template retrieved from the database, this paper determines the candidate area with the highest correlation as the facial area and discriminates whether or not the passport is forged. If the correlation is lower the given threshold, it is determined that the passport may be forged.
- First, the extraction of a skin-color area from the passport image is achieved by using the HSV color model. The features of skin-color areas such as the distributions of color, saturation and intensity are to be calculated heuristically in advance.
- etc. For photo verification, this paper uses various features such as luminance, edge vale, HSV color distribution, RGB color distribution, normalized RGB color distribution, YIQ color distribution and YCbCr color distribution[7-10].
- For the correlation measurement in the photo and face verification, the paper measured various features from the passport images such as: luminance, edge value, HSV color distribution, RGB color distribution, Normalized RGB color distribution, I-value and Q-value of YIQ color model and Cb value and Cr-value of YCbCr color distribution. And in this paper, the features were declared as Luminance, Edge, Hue, RGB, RGB-N, YIQ_I, YIQ_Q, YCbCr-Cb and YCbCr-Cr, respectively.
- The passport images used in this experiment were scanned at 300 DPI by a HP ScanJet 6350C scanner. For the experiment of the photo and face verification, we created the 16 forged passport images in which the whole photo was replaced with another photo and the other 16 forged images with synthesized facial areas. Also, as shown in Fig.
- After compensating heuristically the recognition result, the discrimination of forged passports is achieved by using the photo and face verifications between the photo extracted from the passport image and the picture retrieved from the database based on the recognized codes. In photo and face verification, the correlation was used as the verification measurement and it was measured using Luminance, Edge value, RGB color distribution for the photo verification, and using Hue, YIQ-I value, YCbCr-Cb value for the face verification. When the correlation threshold of 0.
- For example, in the case that the photo in the passport is replaced with another photo, only the automatic code recognition cannot discriminate the passport forgery. So, this paper proposes the whole photo verification method for the discrimination of passport forgery, which extracts the passport photo area from the passport image and verifies it by measuring the similarity between the extracted passport photo and the passport photo image retrieved from the passport database.
- And in this paper, the features were declared as Luminance, Edge, Hue, RGB, RGB-N, YIQ_I, YIQ_Q, YCbCr-Cb and YCbCr-Cr, respectively. The experiment for photo and face verification measured and compared the correlation between the 16 passport photos saved in the database and the 112 photos extracted from forged and varied passport images.
- And, using the code information recognized, the passport photo and related infbnnation of the passport owner is retrieved from the immigration management database. The proposed system measured the passport photo similarity to the face verification by using the correlation that considers totally the overall intensity, edge values and the color information, etc., and it performed the verification separately in the overall picture area and in the facial area to get higher reliability in the discrimination of passport forgery. The overall structure of the proposed system is the same asFig.
- face verification. The proposed system, firstly, extracts and deskewes the areas of passport codes from the passport image by applying the edge detection algorithm and the horizontal minimum value filter to the image. And for the recognition of passport codes, we proposed the new ART algorithm creating the adaptive clusters to the variations of input patterns.
- This paper proposes a novel intelligent passport recognition system, which is able to discriminate passport forgery by using the automatic recognition of passport codes, the passport photo and face verification. The proposed system, firstly, extracts and deskewes the areas of passport codes from the passport image by applying the edge detection algorithm and the horizontal minimum value filter to the image.
- This paper proposes an intelligent immigration control system that authorizes the traveler immigration and detects forged passports by using automatic recognition of passport codes and the photo and face verification. The proposed system extracts and deskewes the code areas from passport images by using properties of passport code areas.
- And, the passport image is deskewed by rotating the pixels as much as the angle. This paper refined the coarse strokes of the code characters by using a 3*3 median filter to prevent the decline of the recognition rate on the low definition images.
- This paper selected the code area with the longest width among the extracted code areas and calculated the angle between the line connected between the centers of the left and the right sides of the code area and the horizontal line in the image. And, the passport image is deskewed by rotating the pixels as much as the angle.
- This paper uses correlation[6] as the similarity measurement between the photo area extracted from the passport image and the photo image retrieved from the passport database. Generally, the Hamming distance or the Euclidean distance is used as the similarity measurement between two vectors.
- This paper, for the improvement of processing performance, applied the combination of a few features having the superior discrimination to the correlation measurement.
- To improve the recognition rate of passport codes with regular appearance, this paper proposed the new ART algorithm that creates the adaptive clusters to the variations of input patterns. The proposed ART learning algorithm is as follows:
대상 데이터
- In the experiment for passport code recognition, 888 code characters with various sizes were extracted from 16 passport images originally scanned and normalized vectors with the size of 20x20 were applied to the learning algorithm so that 344 clusters were created in the proposed ART algorithm. The passport images used in this experiment were scanned at 300 DPI by a HP ScanJet 6350C scanner. For the experiment of the photo and face verification, we created the 16 forged passport images in which the whole photo was replaced with another photo and the other 16 forged images with synthesized facial areas.
이론/모형
- After applying the edge detection algorithm and the horizontal minimum-value filter to the passport image, the output image is binarized by using the iterative thresholding method. The connected objects are extracted from the binarized image by the connected component labeling method[4].
- The connected objects are extracted from the binarized image by the connected component labeling method[4]. By eliminating the useless objects based on the characteristics that the code area is arranged in a row at the bottom of the passport image and is separated from the boundaries of the image, the passport code areas are extracted from the passport image.
성능/효과
- In photo and face verification, the correlation was used as the verification measurement and it was measured using Luminance, Edge value, RGB color distribution for the photo verification, and using Hue, YIQ-I value, YCbCr-Cb value for the face verification. When the correlation threshold of 0.8 was given, the verification experiment on the forged and varied passport images showed a relatively good performance with FAR(False Accept rate) of 3.1% and FRR(False Reject rate) of 2.7%.
후속연구
- Finally, as part of our future work, we plan to implement the recognition and complementary usage of the text information provided on the right-upper of passport image. Also for improvements of flexibility to the site environment, we plan to implement face verification on the actual face image entered directly through the CCD camera at the immigration gate.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.