M. Bichsel. Strategies of robust object recognition for the automatic identification of human faces. PhD thesis, ETH Zurich, 1991. #9467.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....is the representation of the image. We can categorize the image representation into two types: the model based system and the appearance based system. The model based approach uses manually defined features to represent objects in the images. A lot of efforts has been made on this paradigm [2] [5] 6] The focus of this approach is to design an efficient algorithm from a set of manually selected features. The strength of the method is the efficiency in representing images. With a proper design and a restricted domain of images, only a very small number of parameters is sufficient to ....

Martin Bichsel. Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. PhD thesis, Eidgenossischen Technischen Hochschule Zurich, 1991. Diss. ETH Number 9467.

....output is then thresholded for a matching decision. While the class of all face patterns is probably too varied to be modeled by fixed correlation templates, there are some face detection approaches that use a bank of several cor relation templates to detect major facial subfeatures in the image [4] [5] At a later stage, the technique infers the presence of faces from the spatial relationships between the detected subfeatures. A very closely related approach to correlation templates is that of view based eigen spaces [9] The approach assumes that the set of all possible face patterns ....

M. Bichsel. Strategies of Robust Objects Recognition for Automatic Identification of Human Faces. PhD thesis, ETH, Zurich, 1991.

....is the representation of the image. We can categorize the content based image retrieval into two types: the model based and the appearance based. The model based approach uses manually defined features to represent objects in the images. A lot of efforts has been made in this approach [1] [2], 3] 4] 5] Most of them have been focusing on designing an efficient algorithm from a set of manually selected features. The strength of the model based approach is the efficiency in representing images. With a proper design and a restricted domain of images, only a very small number of ....

M. Bichsel, Strategies of Robust Object Recognition for the Automatic Identification of Human Faces, doctoral thesis, Eidgeno ssischen Technischen Hochschule Zu rich, no. 9,467, 1991.

....center of sensed image. In the second task, called pre reaching task, we trained the SAIL robot to reach for the object once it has been located and the eyes fixate on it. Existing studies on visual attention selection are typically based on low level saliency measures, such as edges and texture. [1] In Birnbaum s work [2] the visual attention is based on the need to explore geometrical structure in the scene. In our case, the visual attention selection is a result of past learning experience. Thus, we do not need to define any task specific saliency features. It is the SAIL robot that ....

Martin Bichsel. Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. Swiss Federal Institute of Technology, Zurich, Switzerland, 1991.

....database is the representation of the image. We can categorize the content based image retrieval into two types: the model based and the appearancebased. The model based approach uses manually de ned features to represent objects in the images. A lot of e orts has been made in this approach [1] [2] [3] 4] 5] Most of them have been focusing on designing an ecient algorithm from a set of manually selected features. The strength of the model based approach is the eciency in representing images. With a proper design and a restricted domain of images, only a very small number of parameters is ....

Martin Bichsel, Strategies of Robust Object Recognition for the Automatic Identication of Human Faces, Ph.D. thesis, Eidgenossischen Technischen Hochschule Zurich, 1991, Diss. ETH Number 9467.

....is the representation of the image. We can categorize the image representation into two types: the model based system and the appearance based system. The model based approach uses manually defined features to represent objects in the images. A lot of efforts has been made on this paradigm [2] [5] 6] The focus of this approach is to design an efficient algorithm from a set of manually selected features. The strength of the method is the efficiency in representing images. With a proper design and a restricted domain of images, only a very small number of parameters is sufficient to ....

Martin Bichsel. Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. PhD thesis, Eidgenossischen Technischen Hochschule Zurich, 1991. Diss. ETH Number 9467.

....region and the problem is overcome. 3 Autonomous Learning for Visual Attention Selection The objective of visual attention is to find objects of interest in a scene. Existing studies on visual attention selection are typically based on low level saliency measures, such as edges and texture [2]. In Birnbaum s work [3] the visual attention is based on the need to explore geometrical structure in the scene. In our case, the visual attention selection incorporates the prediction of the critical point of the learned object from the current un centered views. Such a prediction is a result ....

Martin Bichsel. Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. Swiss Federal Institute of Technology, Zurich, Switzerland, 1991.

....the center of sensed image. In the second task, called reaching task, we trained the SAIL robot to reach for the object once it has been located and the eyes fixate on it. Existing studies on visual attention selection are typically based on low level saliency measures, such as edges and texture [1]. In Birnbaum s work [2] the visual attention is based on the need to explore geometrical structure in the scene. In our case, the visual attention selection is a result of past learning experience. Thus, we do not need to de 20 10 0 10 20 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Degree ....

Martin Bichsel. Strategies of Robust Object Recognition for the Automatic Identification of Human Faces.SwissFederal Institute of Technology, Zurich, Switzerland, 1991.

....and census systems to humancomputer interfaces. Human face detection is difficult because there can be huge and unpredictable variations in the appearance of face patterns. Because many of these variations are difficult to parameterize, traditional correlation template pattern matching techniques [1] [2] and geometric model based object recognition approaches tend to perform inadequately for detecting faces. Some nonparametric approaches, such as view based eigen spaces [3] and image invariance schemes [4] have been recently proposed for detecting face patterns, but so far, they have only ....

M. Bichsel. Strategies of Robust Objects Recognition for Automatic Identification of Human Faces. PhD thesis, ETH, Zurich, 1991.

....show the same hand gesture under two different lighting conditions, illustrating that pixel intensities can be sensitive to changes of scene lighting. A pixel by pixel difference of the images (a) and (b) would show a large distance between these identical gestures. However, others have observed [3] that local orientation measurments are less sensitive image T training set feature vector recognition system compare Figure 1: Outline of the recognition system. We apply some transformation T to the image data to form a feature vector which represents that particular gesture. To classify the ....

....to allow more differentiation between gestures. The resulting algorithm is simple and fast to compute. The representation for each gesture is small, and comparisons between feature vectors can be very fast. Our approach relates to other examples of image analysis through orientation analysis. In [3], Bichsel analyzed faces, using local orientation to achieve some lighting invariance. Gorkani and Picard [18] used orientation histograms to compute dominant texture orientations. Nelson [17] used orientation patterns for visual homing. This work is also in the same spirit as texture analysis ....

M. Bichsel. Strategies of robust object recognition for the automatic identification of human faces. PhD thesis, ETH Zurich, 1991. #9467.

....the literature will reveal that OTC as described above is not a new idea, however. Freeman and Roth [18] and Freeman and Weissman [19] make use of essentially the same technique for the task of locating This is page v Printer: Opaque this a human hand in front of a white background, and Bichsel [4] employs dot products between orientation maps for purposes of face recognition. Bichsel in particular provides a large platform of support both computational and biological for the use of orientation maps for feature detection. He cites evidence that measures of local orientation are ....

....of orientation maps for feature detection. He cites evidence that measures of local orientation are robust to the most important object transformations (aside from global rotation) which include both local and global illumination brightness changes and minor variations in object size and shape [4]. We will now discuss our approach to the estimation of local orientation, which is essentially equivalent to the method of Kass Witkin, as described in [26] The first step is to smooth the image with an isotropic filter such as a Gaussian or a difference of Gaussians. The smoothed image, C(x; ....

M. Bichsel. Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. PhD thesis, ETH Zurich, 1991.

....locations, and the output is thresholded for matches. While the class of all face patterns is probably too varied to be modeled by fixed correlation templates, there are some face detection approaches that use a bank of several correlation templates to detect major facial subfeatures in the image [2] [3] The assumption here is that the degree of non rigidity in these facial subfeatures is small enough to be adequately described by a few fixed templates. At a later stage, the technique infers the presence of faces by analyzing the spatial relationship between the detected subfeatures. A ....

M. Bichsel. Strategies of Robust Objects Recognition for Automatic Identification of Human Faces. PhD thesis, ETH, Zurich, 1991.

....nostrils) and then capture the spatial configuration in feature vector whose dimensions typically include measurements like distances, angles, and curvatures. Pictorial approaches, representing faces by using filtered images of model faces, include template based systems ( 2] 6] 13] 7] and [5]) systems using principal components analysis to derive a pictorial face space ( 15] 20] 1] 9] and connectionist approaches ( 16] 11] 10] 21] and [12] 18] explores an interesting hybrid representation that combines the geometrical and pictorial approaches, representing faces ....

Martin Bichsel. Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. PhD thesis, ETH, Zurich, 1991.

....supported by a Howard Hughes Doctoral Fellowship from the Hughes Aircraft Company. 1 Introduction Existing work in face recognition has demonstrated good recognition performance on frontal, expressionless views of faces with controlled lighting (see Baron [4] Turk and Pentland [48] Bichsel [11], Brunelli and Poggio [14] and Gilbert and Yang [20] One of the key remaining problems in face recognition is to handle the variability in appearance due to changes in pose, expression, and lighting conditions. There has been some recent work in this direction, such as pose invariant ....

Martin Bichsel. Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. PhD thesis, ETH, Zurich, 1991.

....contained in the images of the database. As such, content based image retrieval is fundamentally an object recognition problem. The research emphasis to this end has historically been on the design of efficient matching algorithms from a manually designed feature set with hand crafted shape rules [2] [6] 10] 11] 12] 17] 18] 28] Handcrafted shape rules can exploit the efficiency found in manually tuning features for a particular training image set. However, these rules have a severe limitation on the type of object classes that can be found by the image retrieval system. Objects greatly ....

M. Bichsel, Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. PhD thesis, Eidgenossischen Technischen Hochschule Zurich, 1991.Diss. ETH Number 9467.

....Sun Ultra Sparc 2) for the 392 Theta 272 image in figure 2. Other methods are based on first extracting interest points, especially distinguished facial features, such as an elliptical outline [12] the eyes and mouth [9, 18, 19] and local extrema [11, 10] Key features are also prominent in [4, 6, 20]. Efficient focusing is our primary objective. However, in contrast to the work cited above, our approach to visual selection does not utilize complex features, which might be as difficult to detect as the face itself. Instead, we use shape information derived from local primitives, basically edge ....

....flagged locations is based on texture information: After registration and standardization, the greyscale values serve as queries for constructing multiple decision trees. Other proposals for combining edge and texture information appear in [8] and [17] Another link with some prior work, notably [4, 6, 8, 19], as well as our own previous work on shape recognition [2, 3] and model registration ( 1] is the emphasis on geometrical relationships among selected points. In our framework, it is not the points themselves which are distinguished, but rather the global arrangements among their locations. ....

M. Bichsel. Strategies of robust object recognition for the automatic identification of human faces. Technical report, ETH-Zurich, 1991.

....there is sufficient evidence to predict the position of that micro feature. The post processing of the network responses, based on a probabilistic method, is explained in more detail in section 4. 2.3. The use of Local Orientation Information Since neural networks resemble matched filtering [1], low matches will be produced if the pattern presented differs from the patterns learned. Because the gray values in the images depend on lighting conditions, the patterns learned should cover all possible lighting conditions. Hence, instead of gray values we should use image characteristics ....

....all possible lighting conditions. Hence, instead of gray values we should use image characteristics which suffer less from these variations. From various image representations, we have chosen local orientation because it has been proven to be the most robust image property for object recognition [1]. Local orientation information represents the magnitude and the orientation of the intensity gradient. Bichsel [1] has shown that local orientation information is almost invariant to the following object transformations: i) Variations in global contrast, due to variations in the illumination or ....

[Article contains additional citation context not shown here]

M. Bichsel. Strategies of Robust Object Recognition for the Automatic Identifica tion of Human Faces. Ph.d. thesis, ETH Zurich, Zurich, July 16 1991.

....images all are asking the research community for better ways of finding images in large databases. The research emphasis in content based image retrieval has historically been on the design of efficient matching algorithms from a manually created feature set with hand crafted shape rules (e.g. [1] [5] 6] Though these techniques provide efficient matching algorithms, they are typically not able to be generalized to work well on image, object, or database structures other than those for which they were specifically designed. The features which were painstakingly developed for rapid ....

M. Bichsel, Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. PhD thesis, Eidgenossischen Technischen Hochschule Zurich, 1991.Diss. ETH Number 9467.

....analysis, discriminant analysis, hierarchical image database, image retrieval, tessellation, partitioning, complexity with large image databases. 1 Introduction A central task in computer vision module is the recognition of objects from various images of the environment where the machine is found [5] [8] 9] 16] 18] Model based object recognition is the domain where a model exists for every object in the recognition system s universe of discourse. The research emphasis in this paradigm has historically been on the design of efficient matching algorithms from a manually designed feature set ....

....object recognition is the domain where a model exists for every object in the recognition system s universe of discourse. The research emphasis in this paradigm has historically been on the design of efficient matching algorithms from a manually designed feature set with hand crafted shape rules [5] [10] 14] 15] 16] 23] 26] 39] Manually designing a feature set is appealing because such a feature set is very efficient. When designed properly, a very small number of parameters for each of the objects is sufficient to capture the distinguishing characteristics among the objects to be ....

M. Bichsel, Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. PhD thesis, Eidgenossischen Technischen Hochschule Zurich, 1991. Diss. ETH Number 9467.

....contained in the images of the database. As such, content based image retrieval is fundamentally an object recognition problem. The research emphasis to this end has historically been on the design of efficient matching algorithms from a manually designed feature set with hand crafted shape rules [2] [4] 7] 8] 9] 14] 15] 22] Hand crafted shape rules can exploit the efficiency found in manually tuning features for a particular training image set. However, these rules have a severe limitation on the type of object classes that can be found by the image retrieval system. Objects greatly ....

M. Bichsel, Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. PhD thesis, Eidgenossischen Technischen Hochschule Zurich, 1991. Diss. ETH Number 9467.

....Terzopoulos and Waters [37] have used the active contour model of snakes to track facial features in image sequences. In the pictorial approach, a pixel based representation of facial features is matched against the image. This representation may be templates of the major facial features (Bichsel[6], Baron[3] Burt[8] Poggio and Brunelli[7] or the weights of hidden layer nodes in neural networks (Vincent, Waite and Myers[39] For the template based systems, correlation on preprocessed versions of the image is the typical matching metric. The neural network approaches construct a network ....

....images of model faces. In template based systems, the simplest pictorial representation, faces are represented either by images of the whole face or by subimages of the major facial features such as the eyes, nose, and mouth (Baron[3] Brunelli and Poggio[7] Yang and Gilbert[42] Burt[8] Bichsel[6]) Template images need not be taken from the original grey levels; some systems use the gradient magnitude or gradient vector field in order to get invariance to lighting. An input face is then recognized by comparing it to all of the model templates, typically using correlation as an image ....

[Article contains additional citation context not shown here]

Martin Bichsel. Strategies of Robust Object Recognition for the Automatic Identification of Human Faces. PhD thesis, ETH, Zurich, 1991.

....detection and recognition, especially in the processing of mug shots, i.e. head on face images with controlled illumination and scale. The best results have been obtained for 2 D, view based techniques based on either template matching (e.g. 3] combined feature and template matching (e.g. [4]) combined feature and deformable template matching (e.g. 5] minimum distance classification combined with HyberBF interpolation (e.g. 6] or matching using Eigenfaces, i.e. template matching using the Karhunen Loeve transformation of a set of face images (e.g. 7, 1, 8, 9] For mug ....

M. Bichsel, Strategies of Robust Object Recognition for the Automatic Identification of Human Faces, PhD. thesis ETH Zurich, No. 9467, 1991.

No context found.

M. Bichsel. Strategies of robust object recognition for the automatic identification of human faces. PhD thesis, ETH Zurich, 1991. #9467.

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