Matthias Richter

Visual words for automated visual inspection of bulk materials

The inspection of bulk materials in mining, recycling and food-safety places strong requirements on the speed, accuracy and flexibility of automated visual inspection systems. State of the art methods utilize complex feature descriptors and off-the-shelve machine learning techniques. These methods achieve highly accurate results, but typically suffer in execution speed. Commercial systems, on the other hand, use simple features and classifiers to achieve great processing speed, but pay by a complicated intialization procedure and suboptimal classification accuracy. In this paper, we propose to bridge the gap between the two extremes by learning high level object representations that can be used with simple classifiers. For that, we adapt the well known bag of visual words method to use dense sampling and primitive features. The resulting descriptors are very fast to compute and invariant to scale and rotation. At the same time, the method is virtually parameter-free. This allows non-experts to initializate and operate sorting systems based on this approach. We evaluate our method on three food inspection applications. In all experiments we achieve highly accurate, sometimes nearly perfect classification. Comparison to a state of the art method shows that our approach is superior, beating it by a large margin.

Optical filter selection for automatic visual inspection

The color of a material is one of the most frequently used features in automated visual inspection systems. While this is sufficient for many “easy” tasks, mixed and organic materials usually require more complex features. Spectral signatures, especially in the near infrared range, have been proven useful in many cases. However, hyperspectral imaging devices are still very costly and too slow to use them in practice. As a work-around, off-the-shelve cameras and optical filters are used to extract few characteristic features from the spectra. Often, these filters are selected by a human expert in a time consuming and error prone process; surprisingly few works are concerned with automatic selection of suitable filters. We approach this problem by stating filter selection as feature selection problem. In contrast to existing techniques that are mainly concerned with filter design, our approach explicitly selects the best out of a large set of given filters. Our method becomes most appealing for use in an industrial setting, when this selection represents (physically) available filters. We show the application of our technique by implementing six different selection strategies and applying each to two real-world sorting problems.

An approach to color-based sorting of bulk materials with automated estimation of system parameters

Abstract In this paper, we present a flexible method for color-based sorting of bulk materials. It is based on semantically meaningful color features that are constructed from a set of training images. First, estimates of color-occurrence frequencies of different materials are derived from the training images and fused into color classes, which are then used to classify individual pixels. An object descriptor is built as count statistic over the color classes appearing in the object image. This descriptor has many advantages: it is compact and very fast to compute, invariant to scale and rotation, has a very clear, intuitive interpretation, and can be used with simple rule-based classifiers. However, tuning the parameters that govern the feature construction process is laborious and requires a lot of experience on part of the system operator. To overcome this shortcoming, we automatically learn the parameters using genetic algorithms. We apply our method to wine grape sorting problems to show that this approach outperforms a human expert. At the same time, it takes considerably less effort on the human part and frees the expert to attend to other tasks. Furthermore, the system allows non-experts to successfully put a sorting machine in operation.

Large scale classification of spectral signatures

Abstract Hyperspectral sensors are becoming cheaper, faster and more readily available. Apart from industry applications, manufacturers push to bring compact devices into the end-consumer market. This development gives rise to many interesting applications such as the identification of counterfeit pharmaceutical products or the classification of food stuffs. These applications require precise models of the underlying classes. However, building these models from expert knowledge is not feasible. In this paper, we propose to use machine learning techniques to infer a model of many classes from an annotated dataset instead. We investigate the use of three popular methods: support vector machines, random forest classifiers and partial least squares. In contrast to similar approaches using support vector machines, we restrict ourselves to the linear formulation and train the classifiers by solving the primal, instead of dual optimization problem. Our experiments on a large dataset show that the support vector machine approach is superior to random forests and partial least squares in classification accuracy as well as training time.

Extending explicit shape regression with mixed feature channels and pose priors

Facial feature detection offers a wide range of applications, e.g. in facial image processing, human computer interaction, consumer electronics, and the entertainment industry. These applications impose two antagonistic key requirements: high processing speed and high detection accuracy. We address both by expanding upon the recently proposed explicit shape regression [1] to (a) allow usage and mixture of different feature channels, and (b) include head pose information to improve detection performance in non-cooperative environments. Using the publicly available “wild” datasets LFW [10] and AFLW [11], we show that using these extensions outperforms the baseline (up to 10% gain in accuracy at 8% IOD) as well as other state-of-the-art methods.

Gaussian Mixture Trees for One Class Classification in Automated Visual Inspection

We present Gaussian mixture trees for density estimation and one class classification. A Gaussian mixture tree is a tree, where each node is associated with a Gaussian component. Each level of the tree provides a refinement of the data description of the level above. We show how this approach is applied to one class classification and how the hierarchical structure is exploited to significantly reduce computation time to make the approach suitable for real time systems. Experiments with synthetic data and data from a visual inspection task show that our approach compares favorably to flat Gaussian mixture models as well as one class support vector machines regarding both predictive performance and computation time.

Knowing when you don't: Bag of visual words with reject option for automatic visual inspection of bulk materials

Visual inspection of bulk material is the thorough optical inspection of streams of granular material to assess their quality or to detect defective objects. Examples are found in mining (discovery of ores), recycling (sorting waste from reusable material) and food safety (detection of pathogens). In these applications, it is generally not feasible or even possible to provide an accurate and exhaustive training set of all the materials that can be encountered during the inspection. Instead, classification has to be performed in an open world setting, i.e., with the option to recognize and reject unknown objects. Despite the practical relevance, prior work on this topic is surprisingly sparse. Here, we present a method to augment bag of visual words object descriptors by an additional unknown word that encodes outliers. The method depends on only few parameters that have a clear interpretation and is suitable for the application in the field. We demonstrate the performance of our approach using two real-world datasets and compare it to a related method. The experiments show that our method significantly outperforms classification with a closed world assumption as well as the related method.