The problem of learning a generalizable model of the visual appearance of humans from video data is of major importance for computing systems interacting naturally with their users and other humans populating their environment. We propose a step towards automatic behavior understanding by integrating principles of Organic Computing into the posture estimation cycle, thereby relegating the need for human intervention while simultaneously raising the level of system autonomy. The system extracts coherent motion from moving upper bodies and autonomously decides about limbs and their possible spatial relationships. The models from many videos are integrated into meta-models, which show good generalization to different individuals, backgrounds, and attire. These models even allow robust interpretation of single video frames, where all temporal continuity is missing.

The already introduced Neural Map provides a structural association for the building blocks of dynamically generated object models. Its learning and recall procedures are built upon the Growing Neural Gas algorithm, which is highly parameterized. The values of these parameters are obtained through a time-consuming empirical approach. In the present work, we evaluate the use of optimization based on Evolutionary Algorithms to simplify this task. This paradigm delves into six different approaches given by the combination of three fitness functions and two starting conditions. The performance of the proposed optimization paradigm is cross-validated with experiments on invariant object categorization and recognition found in literature. The results show that the empirically set parameter values can be successfully optimized using this paradigm.

editorial Free AccessEditorial: Special issue on organic computing Authors: Rolf P. Würtz Ruhr-University Bochum Ruhr-University BochumView Profile , Kirstie L. Bellman The Aerospace Corporation The Aerospace CorporationView Profile , Hartmut Schmeck Karlsruhe Institute of Technology Karlsruhe Institute of TechnologyView Profile , Christian Igel Ruhr-University Bochum Ruhr-University BochumView Profile Authors Info & Claims ACM Transactions on Autonomous and Adaptive SystemsVolume 5Issue 3Article No.: 9pp 1–3https://doi.org/10.1145/1837909.1837910Published:30 September 2010Publication History 1citation369DownloadsMetricsTotal Citations1Total Downloads369Last 12 Months22Last 6 weeks4 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteeReaderPDF

We present a so-called Neural Map, a novel memory framework for visual object recognition and categorization systems. The properties of its computational theory include self-organization and intelligent matching of the image features that are used to build their object models. Its performance for representing the visual object knowledge comprised by these models and for recognizing unknown objects is measured using three different types of image features, which extract different granularity of information from object views of the ETH-80 image set. The obtained experimental results slightly outperform previous ones using PCA-based methods on the same image set, and they suggest that the medium-sized image features maximize the object models' informativeness and distinctiveness.

We present an integrated face recognition system that combines a Maximum Likelihood (ML) estimator with Gabor graphs for face detection under varying scale and in-plane rotation and matching as well as a Bayesian intrapersonal/extrapersonal classifier (BIC) on graph similarities for face recognition. We have tested a variety of similarity functions and achieved verification rates (at FAR 0.1%) of 90.5% on expression-variation and 95.8% on size-varying frontal images within the CAS-PEAL database. Performing Experiment 1 of FRGC ver2.0, the method achieved a verification rate of 72%.

We present an object recognition system built on a combination of feature- and correspondence-based pattern recognizers. The feature-based part, called preselection network, is a single-layer feedforward network weighted with the amount of information contributed by each feature to the decision at hand. For processing arbitrary objects, we employ small, regular graphs whose nodes are attributed with Gabor amplitudes, termed parquet graphs. The preselection network can quickly rule out most irrelevant matches and leaves only the ambiguous cases, so-called model candidates, to be verified by a rudimentary version of elastic graph matching, a standard correspondence-based technique for face and object recognition. According to the model, graphs are constructed that describe the object in the input image well. We report the results of experiments on standard databases for object recognition. The method achieved high recognition rates on identity and pose. Unlike many other models, it can also cope with varying background, multiple objects, and partial occlusion.

The current interest in organic computing is also sustained by a notion of "organic" which relates to the user rather than the developer. In that aspect, organic computing requires that the interface between the IT system and the user be organic, intelligible, and friendly. This again imposes constraints on the user interfaces. On an abstract level, a recognition systems requires a data format for stored objects, a method to compare given images with those objects, and a method to turn the resulting similarities into a decision about the object's identity. For efficient application to real world data, the efficient organization of the object database and methods to presegment images and image sequences become important. The subsystems to perform these tasks can be designed by using the organic computing principles of "learning from nature" and "self-organization".

We present an analysis of the representation of images as the magnitudes of their transform with complex-valued Gabor wavelets. Such a representation is a model for complex cells in the early stage of visual processing and of high technical usefulness for image understanding, because it makes the representation insensitive to small local shifts. We show that if the images are band limited and of zero mean, then reconstruction from the magnitudes is unique up to the sign for almost all images.

We describe an object recognition system based on symbolic contour graphs. The image to be analyzed is transformed into a graph with object corners as vertices and connecting contours as edges. Image corners are determined using a robust multiscale corner detector. Edges are constructed by line-following between corners based on evidence from the multiscale Gabor wavelet transform. Model matching is done by finding subgraph isomorphisms in the image graph. The complexity of the algorithm is reduced by labeling vertices and edges, whereby the choice of labels also makes the recognition system invariant under translation, rotation and scaling. We provide experimental evidence and theoretical arguments that the matching complexity is below O(#V3), and show that the system is competitive with other graph-based matching systems.

The Gestalt principle of collinearity (and curvilinearity) is widely regarded as being mediated by the long-range connection structure in primary visual cortex. We review the neurophysiological and psychophysical literature to argue that these connections are developed from visual experience after birth, relying on coherent object motion. We then present a neural network model that learns these connections in an unsupervised Hebbian fashion with input from real camera sequences. The model uses spatiotemporal retinal filtering, which is very sensitive to changes in the visual input. We show that it is crucial for successful learning to use the correlation of the transient responses instead of the sustained ones. As a consequence, learning works best with video sequences of moving objects. The model addresses a special case of the fundamental question of what represents the necessary a priori knowledge the brain is equipped with at birth so that the self-organized process of structuring by experience can be successful.

As a step towards systems that can acquire knowledge automatically, we have designed a system that can learn new objects with a minimum of user interaction and implemented it on our robot platform GripSee (M. Becker et al., 1999). A novel object is placed into the robot's gripper in order to define a default orientation and a default grip. The robot then places the object on a turning table and builds up a visual representation that consists of a collection of graphs, labeled with multiscale edges. A user interface that can correct errors in the representation is also part of the system. The visual representation is complemented by a grip library, which contains possible ways of grasping and manipulating the object in a robust manner. We regard this procedure as an example of human assisted learning.

Recognition systems based on model matching using low level features often fail due to a variation in background. As a solution, I present a system for the recognition of human faces independent of hairstyle. Correspondence maps between an image and a model are established by coarse-fine matching in a Gabor pyramid. These are used for hierarchical recognition.

An object recognition system based on the dynamic link architecture, an extension to classical artificial neural networks (ANNs), is presented. The dynamic link architecture exploits correlations in the fine-scale temporal structure of cellular signals to group neurons dynamically into higher-order entities. These entities represent a rich structure and can code for high-level objects. To demonstrate the capabilities of the dynamic link architecture, a program was implemented that can recognize human faces and other objects from video images. Memorized objects are represented by sparse graphs, whose vertices are labeled by a multiresolution description in terms of a local power spectrum, and whose edges are labeled by geometrical distance vectors. Object recognition can be formulated as elastic graph matching, which is performed here by stochastic optimization of a matching cost function. The implementation on a transputer network achieved recognition of human faces and office objects from gray-level camera images. The performance of the program is evaluated by a statistical analysis of recognition results from a portrait gallery comprising images of 87 persons.>