C. Kotropoulos and I. Pitas. A rule based face detection in frontal views. In ICASP, volume IV, pages 2537--2540, 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....facial features are applied. Although this method does not result in a high localization rate, the idea of mosaicing, multiple levels and rules to guide the search have influenced other peoples work in this area. In order to avoid the iterative nature of Yang s method [57] Kotropoulos and Pitas [29] propose to estimate the cell dimensions in the quartet image by processing the horizontal and vertical profile (proposed by Kanade [24] obtained by averaging all pixel intensities in each column and row, respectively. The projection method can be e#ective in determining the positions of ....

C. Kotropoulos and I. Pitas. Rule-based face detection in frontal views. In Proc. of International Conference on Acoustics, Speech, and Signal Processing, pages 21--24, 1997. 28

....is a fundamental problem in computer vision. One of the typical and challenging applications in this field that has been extensively investigated is the problem of human face detection and localization. A variety of methods have been developed for solving this problem: Top Down approaches [1,2], Bottom Up approaches [3,4] Statistical approaches [5,6] Neural network approaches [7,8] Fast algorithms [9,10] Template matching by means of matched filter [11] Template matching by means of linear adaptive filters [12] and others methods [13,14] In this paper, we present an approach to ....

C. Kotropoulos and I. Pitas, "Rule-based face detection in frontal views". IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP'99), Phoenix, Arizona, pp. 25372540, 1997.

....of one or more features is imprecise (figure 6) In 91 of the images both the eyes and the mouth are localised within a 3 pixel tolerance, i.e. one tenth of the typical inter eye distance. Results on facial features detection on the M2VTS database by another research group have been published in [5], where an 86:5 success rate for simultaneous detection of eyebrows eyes, nostrils nose and mouth over 37 frontal images from the database is reported. Although the results of our experiments seem to fall somewhat short of the optimal model performance reported in the last column of table 1, one ....

....the results of our experiments seem to fall somewhat short of the optimal model performance reported in the last column of table 1, one must keep in mind that the difficulty of the problem is now considerably increased. This difficulty seems to be confirmed by others using the same database [5]. On one hand, the system is now required to find its own facial feature candidates without exploring the whole image; on the other hand, no particular threshold is adjusted to bring each classifier to its optimal working point; finally, the number of negative examples is now potentially enlarged ....

C. Kotropoulos and I. Pitas. Rule--based face detection in frontal views. In Proceedings of the IEEE International Conference 1 Approximately 2400 frontal images taken at four different periods. on Acoustics, Speech and Signal Processing (ICASSP'97), volume IV, pages 2537--2540, April 1997.

No context found.

C. Kotropoulos and I. Pitas, \Rule-based face detection in frontal views", in Proc. of the IEEE ICASSP '97, April 1997, Munich, Germany, pp. 2537-2540.

....as mosaic images) The algorithm attempts to detect a facial region at a coarse resolution and subsequently to validate the outcome by detecting facial features at the next resolution using a hierarchical knowledge based pattern recognition system. A variant of this method has been proposed in [9] that allows for rectangular cells instead of square cells and provides estimates of the cell dimensions and the offsets so that the mosaic model fits the face image of a person by preprocessing the horizontal and the vertical profile of the image. The original algorithm [8] is based on images of ....

....are created by ordering lexicographically the grey levels of the quartet image cells that fall inside a window scanning the quartet image. Alternatively, one may use the horizontal and vertical image profiles in order to extract a bounding box for the face region, as has been demonstrated in [9]. The horizontal profile of the image is obtained by averaging all pixel intensities in each image column. Similarly, the vertical profile of the image is obtained by averaging all pixel intensities in each image row. Instead of locating the extrema of the aforementioned profiles and defining ....

[Article contains additional citation context not shown here]

C. Kotropoulos and I. Pitas, "Rule-based face detection in frontal views," in Proc. of the 1997 IEEE Int. Conf. on Acoustics, Speech, and Signal Processing, pp. 2537--2540, 1997.

....distance between two nodes of the sparse grid that is to be created. The value oe m = 9 has been used in all experiments reported in this paper. An 8 Theta 8 sparse grid has been created by measuring the feature vectors j(x) at equally spaced nodes over the output of the face detection algorithm [8], a variant of the method proposed by Yang and Huang [7] Fig. 1 depicts the output of multiscale dilation erosion for the various scales used. The first nine pictures starting from the upper left picture are dilated images and the remaining nine are eroded images. It is seen that multiscale ....

....optimization procedure suffices for the minimization of such a cost function [2] The above mentioned approach is proven inadequate in our experiments. October 22, 1999 DRAFT 6 Accordingly, we propose: i) to exploit the face detection outcome provided by the hierarchical rule based system [8] for initializing the minimization of the cost function, and (ii) to replace the two stage optimization procedure by a probabilistic hill climbing algorithm (i.e. a simulated annealing algorithm) that is reminiscent of the Algorithm 1.4 [11, p. 12] that does not make distinction between coarse ....

C. Kotropoulos, and I. Pitas, "Rule-based face detection in frontal views," in Proc. of the IEEE Int. Conf. on Acoust, Speech and Signal Proc.(ICASSP 97), pp. 2537--2540, Munich, Germany 1997.

.... Frontal face detection A very attractive approach for face detection based on multiresolution images (also known as mosaic images) has been proposed in [23] Motivated by the simplicity of this face detection approach, we briefly describe a variant of this method that has the following features [24]: a) It uses rectangular cells in contrast to the square cells used in [23] b) It is equipped with a September 30, 1999 DRAFT 8 preprocessing step that determines an estimate of the cell dimensions and the offsets so that the mosaic model fits the face image of each person. c) It has very low ....

....to 2 Theta 2 cells of half dimensions the octet image results, where the main facial features, such as eyebrows eyes, nostrils nose and mouth, are detected. Due to lack of space, the detailed description of the face detection algorithm employed is omitted and the interested reader is referred to [24]. Figure 3 depicts two frontal face images. Their quartet and octet images are shown in the same figure. The images in the last column are the results of the face detection algorithm. The octets for the facial features are shown overlaid in these images. The octets for eyebrows eyes and ....

C. Kotropoulos, and I. Pitas, "Rule-based face detection in frontal views," in Proc. of the IEEE Int. Conf. on Acoust, Speech and Signal Proc.(ICASSP 97), pp. 2537--2540, Munich, Germany 1997.

....facial features (e.g. nose, eyes, etc. called fiducial points. In both cases a face facial feature detection algorithm is needed. In this paper, we mainly resort to a variant of the approach proposed by Yang and Huang [10] that is based on multiresolution images, the so called mosaic images [11]. Let the superscripts t and r denote a test and a reference person (or grid) respectively. The L 2 norm between the feature vectors at the l th grid node is used as a (signal) similarity measure, i.e. C v (j(x t l ) j(x r l ) kj(x t l ) Gamma j(x r l )k. The objective in ....

C. Kotropoulos and I. Pitas, "Rule-based face detection in frontal views," in Proc. of the IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP97) , Munich, Germany, April 1997, vol. IV, pp. 2537-- 2540.

....facial features (e.g. nose, eyes, etc. called fiducial points. In both cases a face facial feature detection algorithm is needed. In this paper, we mainly resort to a variant of the approach proposed by Yang and Huang [10] that is based on multiresolution images, the so called mosaic images [11]. Let the superscripts t and r denote a test and a reference person (or grid) respectively. The L 2 norm between the feature vectors at the l th grid node is used as a (signal) similarity measure, i.e. C v (j(x t l )# j(x r l ) kj(x t l ) j(x r l )k. The objective in elastic graph ....

C. Kotropoulos and I. Pitas, "Rule-based face detection in frontal views," in Proc. of the IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP97) , Munich, Germany, April 1997, vol. IV, pp. 2537-- 2540.

No context found.

C. Kotropoulos and I. Pitas, Rule-based face detection in frontal views, in Proc. of IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP 97) IV, 2537-2540 (1997).

No context found.

C. Kotropoulos and I. Pitas, Rule-based face detection in frontal views, Proc. ICASSP '97, 2537-2540, (1997).

.... shape and skin colour [1] Another very attractive approach for face detection relies on multiresolution images (also known as mosaic images) attempting to detect a facial region at a coarse resolution and, subsequently, to validate the outcome by detecting facial features at the next resolution [2, 3]. The major difficulties encountered in face recognition are due to variations in luminance, facial expressions, visual angles and other features such as glasses, beard, etc. This leads to a need for employing several rules in the algorithms that are used, in order to tackle with these problems. ....

C. Kotropoulos and I. Pitas, "Rule-based face detection in frontal views", in Proc. of the IEEE ICASSP '97, April 1997, Munich, Germany, pp. 2537-2540.

.... : 9 form the feature vector located at the grid node x: j(x) f g 9 ) x) f(x) f g Gamma9 ) x) 2) An 8 Theta 8 sparse grid has been created by measuring the feature vectors j(x) at equally spaced nodes over the output of the face detection algorithm described in [7]. j(x) has been demonstrated that captures important information for the key facial features [8] Another method for modeling a grayscale facial image region is to employ the morphological signal decomposition (MSD) Let us denote by f(x) D Z 2 Z the facial image region that can be ....

.... for the key facial features [8] Another method for modeling a grayscale facial image region is to employ the morphological signal decomposition (MSD) Let us denote by f(x) D Z 2 Z the facial image region that can be extracted by using a face detection module such as the one proposed in [7]. Without any loss of generality it is assumed that the image pixel values are non negative, i.e. f(x) 0. Let g(x) 1, 8x : kxk oe denote the structuring function. The value oe = 2 has been used in all experiments. Symmetric operators will not explicitly denoted hereafter. Given f(x) and g(x) ....

C. Kotropoulos, and I. Pitas, "Rule-based face detection in frontal views," in Proc. of IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP 97), vol. IV, pp. 2537--2540, Munich, Germany, April 21-24, 1997.

....a facial region at a coarse resolution and subsequently to validate the outcome by detecting facial features at the next resolution level [15] Towards this goal, the method employs a hierarchical knowledge based pattern recognition system. Recently, a variant of this method has been proposed [7]. It offers the following features: a) It allows for rectangular cells in contrast to the square cells used in [15] b) It is equipped with a preprocessing step that determines an estimate of the cell dimensions and the offsets so that the mosaic model fits the face image of each person. c) It ....

C. Kotropoulos and I. Pitas. Rule-based face detection in frontal views. In Proc. of the IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP-97), volume IV, pages 2537--2540, Munich, Germany, April 1997.

....nostrils nose and mouth detection rules are developed to validate the facial candidates determined by the procedure outlined above. The rules developed enhance the ones proposed in [2] by highlighting the key role of symmetry. A detailed description of the rules implemented is given in [12]. The proposed algorithm has been applied to the European ACTS project M2VTS database [11] The database includes the videosequences of 37 different persons. The algorithm provides a correct facial candidate in all cases. However, the detected facial features that validate the choice of the facial ....

....in all cases. However, the detected facial features that validate the choice of the facial candidate are not always correct. The true success rate of the proposed method in detecting simultaneously eyebrows eyes, nostrils nose and mouth is 86.5 under the most strict evaluation conditions [12]. Fig. 2 shows facial candidates that have successfully been detected and the facial features are located correctly. The octets for eyebrows eyes and nostrils nose are shown as white overlaid rectangles. Mouth candidates are shown as black overlaid rectangles. The white cross indicates the ....

Kotropoulos, C., and Pitas, I.: Rule-based face detection in frontal views. In Proc. of the 1997 IEEE Int. Conf. on Acoustics, Speech and Signal Processing (to appear) This article was processed using the L a T E X macro package with LLNCS style

.... is to find the test grid node coordinates fx t i ; i 2 Vg that minimize C(fx t i g) X i 2V 8 : S v (J(x t i ) J(x r i ) X j 2N (i) S e (i; j) 9 = 12) The reference grid (i.e. the model grid) has been placed over the output of face detection algorithm described in [10]. An 8 Theta 8 sparse grid of equally spaced nodes has been employed. The outputs of multiscale dilation erosion for scales oe = Gamma9; 9 have been concatenated to form the feature vector at each grid node. The cost function (12) is actually a matching error that defines a distance ....

....a two stage coarseto fine optimization procedure suffices for the minimization of (12) In our experiments, the above mentioned approach is proved inadequate. Accordingly, we propose: i) to exploit the face detection results that are provided by the hierarchical rule based system described in [10] for initializing the minimization of the cost function, and (ii) to replace the two stage optimization procedure by a probabilistic hill climbing algorithm (i.e. a simulated annealing algorithm) that is reminiscent of the Algorithm 1.4 [5, p. 12] that does not make distinction between coarse and ....

C. Kotropoulos, and I. Pitas, "Rule-based face detection in frontal views," in Proc. of IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP 97), vol. IV, pp. 2537--2540, Munich, Germany, April 21-24, 1997.

....find the test grid node coordinates x t i , i # V that minimize C( x t i ) # i#V # # # Sv (J(x t i ) J(x r i ) # # j#N (i) Se(d t ij , d r ij ) # # # . 8) The reference grid (i.e. the model grid) has been placed over the output of face detection algorithm described in [12]. An 8 8 sparse grid of equally spaced nodes has been employed. The outputs of multiscale dilation erosion for scales # = 9, 9 have been concatenated to form the feature vector at each grid node. The cost function (8) is actually a matching error that defines a distance measure between ....

....that a two stage coarse to fine optimization procedure su#ces for the minimization of (8) In our experiments, the above mentioned approach is proved inadequate. Accordingly, we propose: i) to exploit the face detection results that are provided by the hierarchical rule based system described in [12] for initializing the minimization of the cost function, and (ii) to replace the two stage optimization procedure by a probabilistic hill climbing algorithm (i.e. a simulated annealing algorithm) that is reminiscent of the Algorithm 1.4 [7, p. 12] that does not make distinction between coarse and ....

C. Kotropoulos, and I. Pitas, "Rule-based face detection in frontal views," in Proc. of IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP 97), vol. IV, pp. 2537--2540, Munich, Germany, April 21-24, 1997.

.... at the grid node x: j(x) f g 9 ) x) f g 1 ) x) f(x) f g Gamma1 ) x) f g Gamma9 ) x) 3) An 8 Theta 8 sparse grid has been created by measuring the feature vectors j(x) at equally spaced nodes over the output of the face detection algorithm described in [10]. It has been demonstrated that such feature vectors captures important information for the key facial features [5] Subsequently, a dimensionality reduction of feature vectors is pursued by employing PCA. PCA methods have shown good performance in image reconstruction compression tasks. ....

C. Kotropoulos, and I. Pitas, "Rule-based face detection in frontal views," in Proc. of IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP 97), vol. IV, pp. 2537--2540, Munich, Germany, April 21-24, 1997.

.... color that corresponds to a particular sector of the HSV color space and their elliptical shape [22] When only grey level information is available, face detection can be based on a hierarchical knowledge based pattern recognition system that uses multiresolution images, known as mosaic images [23, 24]. Both approaches employ a number of assumptions, including the mirrorsymmetry of the face, the biometric analogies (i.e. a prescribed ratio of face height over face width) the fact that key facial features, such as eyebrows, eyes, nostrils, etc. are related to image local minima identifiable ....

C. Kotropoulos, and I. Pitas, "Rule-based face detection in frontal views," in Proc. of IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP 97), vol. IV, pp. 2537--2540, 1997.

.... : 9 form the feature vector located at the grid node x: j(x) f g9) x) f(x) f g Gamma9 ) x) 2) An 8 Theta 8 sparse grid has been created by measuring the feature vectors j(x) at equally spaced nodes over the output of the face detection algorithm described in [4]. j(x) has been demonstrated that captures important information for the key facial features [5] Subsequently, a feature vector dimensionality reduction is pursued by employing PCA. In addition to dimensionality reduction PCA decorrelates the feature vectors and facilitates the LDA that is ....

....LINK ARCHITECTURE The modeling of a gray scale facial image region by employing morphological shape decomposition (MSD) is described in this section. Let us denote by f(x) D Z 2 Z the facial image region that can be extracted by using a face detection module such as the one proposed in [4]. Without any loss of generality it is assumed that the image pixel values are non negative, i.e. f(x) 0. Let g(x) 1, 8x : kxk oe denote the structuring function. The value oe = 2 has been used in all experiments. Symmetric operators will not explicitly denoted hereafter. Given f(x) and ....

C. Kotropoulos, and I. Pitas, "Rule-based face detection in frontal views," in Proc. of IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP 97), vol. IV, pp. 2537--2540, Munich, Germany, April 21-24, 1997.

.... form the feature vector located at the grid node x, i.e. j(x) f g9) x) f(x) f g Gamma9 ) x) An 8 Theta 8 sparse grid has been created by measuring the feature vectors j(x) at equally spaced nodes over the output of the face detection algorithm described in [10]. For a complete description of MDLA, the interested reader may refer to [4] MDLA was tested on the XM2VTSDB according to the two configurations defined in [9] A subset of 200 persons (i.e. training clients) is used for training. Let us denote by S1 the set of training clients. The design of ....

C. Kotropoulos, and I. Pitas, "Rule-based face detection in frontal views," in Proc. of IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP 97), vol. IV, pp. 2537--2540, Munich, Germany, April 21-24, 1997.

No context found.

C. Kotropoulos and I. Pitas. A rule based face detection in frontal views. In ICASP, volume IV, pages 2537--2540, 1997.

No context found.

C. Kotropoulos, I. Pitas. Rule -- Based Face Detection in Frontal Views. In Proceedings of International Conference on Acoustics, Speech, and Signal Processing (ICASSP'99). pp. 21-24. Phoenix, Arizona, 1997.

No context found.

C. Kotropoulos and I. Pitas. Rule-based face detection in frontal views. In Proceedings in Interantional Conference on Acoustics, speech and signal processing, 1997.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC