Unsupervised shape discovery using synchronized spectral networks

We consider the problem of object recognition via local geometric feature matching in the presence of sensor uncertainty, occlusion, and clutter. We present a general formulation of the problem and a polynomial-time algorithm which guarantees finding all geometrically feasible interpretations of the data, modulo uncertainty, in terms of the model. This formulation applies naturally to problems involving both 2D and 3D objects.The primary contributions of this work are the presentation of a robust, provably correct, polynomial-time approach to this class of recognition problems and a demonstration of its practical application; and the development of a general framework for understanding the fundamental nature of the geometric feature matching problem. This framework provides insights for analyzing and improving previously proposed recognition approaches, and enables the development of new algorithms.

Object recognition is of fundamental importance in computer vision. In a few years, pedestrian detection, car detection, and more generally scene recognition will likely be reliable enough to allow fully-automated car navigation, and the human driver will be relegated to the back seat to sip his coffee. In this thesis we are interested in recognizing individual objects and categories. In order to reduce the volume of information one has to process, images are characterized by sets of features. These features, also called interest points, are targeted at image locations with high local information content. Various systems for detecting interest points and for describing the local image appearance near these points, have been proposed in the last two decades. We investigate which combinations from this plethora of detectors and descriptors, are most suited for object recognition tasks. On to the problem of object recognition, we are first interested in recognizing individual objects. In a few years, one can imagine that customers in shops, will take with their cell phone a picture of a product that looks interesting, send it to a remote server with a huge database of individual objects, and get back information about that specific product. We propose a system for individual object recognition, inspired from previous work on coarse-to-fine recognition. All steps of the recognition process are translated into principled probabilistic terms, which allows us to outperform a state-of-the-art commercial system for individual recognition. Regarding categories, faces are probably the category that has received the most attention in computer vision literature. Here we propose a system to recognize images of the same individual in large databases of images. This can be of high interest when looking for images of a given person over the internet. Our method's advantage is that it works on real-world images, as opposed to the face databases from the literature, collected in laboratories with controlled lighting, pose and background conditions. Finally, we are interested in recognition of object categories in general. Using support vector machines for the classification task, we propose a features-based kernel that improves recognition performance on object categories.

In Image Processing, the problem of Object Recognition is regarded as difficult. It's an important aspect of PC vision because so many PC vision programmes rely on it for effectiveness. To address this issue, a number of identifying terminology and frameworks have been discussed for quite some time. The purpose of this paper is to give a briefoverview of a range of platforms in the field of computer vision and object recognition. For the most part this paper covers a relatively mature service from AWS(Amazon Web Services), Amazon Rekognition which is easy to integrate into other services like Lambda, providing huge volume of details in its analysis. The paper incorporates different viewpoints that have been utilized by various analysts for object spotting and recognition. Rekognition is able to recognise objects and scenarios within photographs in this article, returning a confidence measure ranging from 0 to 100% for each label returned. It accomplishes this by using AWS’ Python client Boto3 and the detect labels method. This works by supplying an image, a limit on the number of labels to return, and a confidence level for filtering out ambiguous results.

3-D object recognition is a process of matching an object to a scene description to determine the object’s identity and/or its pose (position and orientation) in space. The problem of object recognition can be separated into two closely related subproblems — that of model building and that of recognition. Object recognition using multiple-view approach is achieved by finding a best match between the description of object projection and one of the model descriptions in the database. An object is modeled by its 2-D projections as seen from a set of predetermined viewpoints uniformly distributed on a view sphere enclosing the object. In our experiments, an icosahedron (twenty-sided) is used to approximate the view sphere. The method to search the model database is crucial to the performance of the recognition process. To recognize an object, a coarse-to-fine search strategy is adopted to reduce the processing time. At the coarse search stage, a small set of views most similar to the object in the scene is located. In most cases, this step vastly decreases the search space. The fine (or exhaustive) search process starts with these located views and compares the topology of the object in the scene with the projections rendered by the neighboring viewpoints of the located viewpoints until a best match is found. In this paper, a FKCN model which is an integration of Fuzzy c-Means (FCM) and Kohonen clustering networks (KCNs) is proposed to exploit the parallelism at the coarse search stagein object recognition, where the learning rate for Kohonen updating is proposed to be: $$\alpha _{ik,t}= (u_{ik,t} )^m t$$ where \(\alpha _{ik,t}= \left( {\sum {\left( {\parallel x_k- v_i \parallel _A /\parallel _{x_k }- v_j \parallel _A } \right)^{2/(m - 1)} } } \right)^{ - 1}\) by minimizing \(\sum\limits_i {\sum\limits_{\mathfrak{k}^u ik} m } \left( {\parallel x_k- v_i \parallel _A } \right)^2\),but weighting exponent m is a function of t, the iterate number, i.e. m = mt = (m0+tΔm) (an increasing strategy). Feature vectors are first extracted from each view. Then a FKCN clusters these feature vectors to get cluster centers, each of which is regarded as a prototype of some similar views. Finally, a feature vector from an input view is compared to the cluster centers instead of all views and finds the potential candidates which are clustered into the closest cluster center in order to achieve the goal of reducing search space.

Model-based object recognition has been a very active research field in Computerter Vision during the past ten years. Several excellent survey papers have been published, for example, [10.1,2]. Crimson has published a book [10.3] describing an extended series of experiments, carried out by himself and his colleagues at MIT, on the role of geometry in solving the object recognition and localization problem. Their approach is mainly based on the search in the interpretation tree, as briefly presented in Sect. 6.1. One chapter of the book by Faugeras [10.4] is also devoted to this problem based in detail with this immense subject. We want simply to adapt our motion-determination algorithm developed in Chap. 7 to solve some primitive object recognition problems.

Comparative perspectives on multiple cortical visual systems

Active/space-variant object recognition

Robust scene recognition using a GA and real-world raw-image

This research investigates fundamental problems in object recognition in earthen heritage and addresses the possibility of an automatic crack detection method for rammed earth images. We propose and validate a straightforward support vector machine (SVM)-based bidirectional morphological approach to automatically generate crack and texture line maps through transforming a surface image into an intermediate representation. Rather than relying on the application of the eight connectivity rule to a combination of horizontal and vertical gradient to extract edges, we instruct an edge classifier in the form of a support vector machine from features computed on each direction separately. The model couples a bidirectional local gradient and geometrical characteristics. It constitutes of four elements: (1) bidirectional edge maps; (2) bidirectional equivalent connected component maps; (3) SVM-based classifier and (4) crack and architectural line feature map generation. Relevant details are discussed in each part. Finally, the efficiency of the proposed algorithm is verified in a set of simulations that is satisfactorily conforming to labeled data provided manually for surface images of earthen heritage.

In nature, species showcase a spectrum of unique behaviors, a trait promising for emulation in intelligent agents. Existing unsupervised skill discovery methods often rely on single-skill conditioned policies, constraining the breadth of discovered skills. In this paper, we introduce a novel approach to unsupervised skill discovery, termed Speciated Skill Discovery (SSD), which trains a speciated population of skill-conditioned policies. SSD maximizes the mutual information between states and skills and between state and species-given skills. To achieve this, we employ a contrastive learning framework to minimize the conditional entropy between states and skill-species pairs, facilitating the learning of controllable latent behaviors. Moreover, we utilize a particle-based entropy estimator to maximize state entropy, thereby promoting state space exploration. Our experimental results demonstrate that SSD uncovers a range of innovative skills, surpassing the performance of prior unsupervised skill discovery methods.

Small target detection based on automatic ROI extraction and local directional gray&entropy contrast map

Machine face recognition has traditionally been studied under the assumption of a carefully controlled image acquisition process. By controlling image acquisition, variation due to factors such as pose, lighting, and background can be either largely eliminated or specifically limited to a study over a discrete number of possibilities. Applications of face recognition have had mixed success when deployed in conditions where the assumption of controlled image acquisition no longer holds. This dissertation focuses on this unconstrained face recognition problem, where face images exhibit the same amount of variability that one would encounter in everyday life. We formalize unconstrained face recognition as a binary pair matching problem (verification), and present a data set for benchmarking performance on the unconstrained face verification task. We observe that it is comparatively much easier to obtain many examples of unlabeled face images than face images that have been labeled with identity or other higher level information, such as the position of the eyes and other facial features. We thus focus on improving unconstrained face verification by leveraging the information present in this source of weakly supervised data. We first show how unlabeled face images can be used to perform unsupervised face alignment, thereby reducing variability in pose and improving verification accuracy. Next, we demonstrate how deep learning can be used to perform unsupervised feature discovery, providing additional image representations that can be combined with representations from standard hand-crafted image descriptors, to further improve recognition performance. Finally, we combine unsupervised feature learning with joint face alignment, leading to an unsupervised alignment system that achieves gains in recognition performance matching that achieved by supervised alignment.

Unsupervised discovery of sub-lexical units in speech is a problem that currently interests speech researchers. In this paper, we report experiments in which we use phone segmentation followed by clustering the segments together using k-means and a Convolutional Neural Network. We thus obtain an annotation of the corpus in pseudo-phones, which then allows us to find pseudo-words. We compare the results for two different segmentations: manual and automatic. To check the portability of our approach, we compare the results for three different languages (English, French and Xitsonga). The originality of our work lies in the use of neural networks in an unsupervised way that differ from the common method for unsupervised speech unit discovery based on auto-encoders. With the Xitsonga corpus, for instance, with manual and automatic segmentations, we were able to obtain 46% and 42% purity scores, respectively, at phone-level with 30 pseudo-phones. Based on the inferred pseudo-phones, we discovered about 200 pseudo-words.

Our visual world contains numerous objects. An important function of the visual system is to recognize an object by comparing perceptual and memory representations. Although we seldom have any problems in recognizing natural objects, which promotes the belief that object recognition is a quite simple process of matching perceptual and memory representations, recognition does in fact involve extremely complicated visual processing. The fact that objects are almost never presented in isolation illustrates the complexity and difficulty of object recognition. The cluttered nature of our visual environment poses an object segmentation problem (including figure/ground segregation problem), which itself is quite difficult. Even if one can successfully segment a set of objects, there is another problem for the visual system to solve: the so-called binding problem. If there are multiple objects, each of which has its own feature values such as shape, color, size, and so on, then how does the visual system properly maintain the correct correspondences of these features? This chapter focuses on this binding problem in both object recognition and visual working memory.

The article deals with the problems of motion detection, object recognition, and scene description using deep learning in the framework of granular computing and Z-numbers. Since deep learning is computationally intensive, whereas granular computing, on the other hand, leads to computation gain, a judicious integration of their merits is made so as to make the learning mechanism computationally efficient. Further, it is shown how the concept of z-numbers can be used to quantify the abstraction of semantic information in interpreting a scene, where subjectivity is of major concern, through recognition of its constituting objects. The system, thus developed, involves recognition of both static objects in the background and moving objects in foreground separately. Rough set theoretic granular computing is adopted where rough lower and upper approximations are used in defining object and background models. During deep learning, instead of scanning the entire image pixel by pixel in the convolution layer, we scan only the representative pixel of each granule. This results in a significant gain in computation time. Arbitrary-shaped and sized granules, as expected, perform better than regular-shaped rectangular granules or fixed-sized granules. The method of tracking is able to deal efficiently with various challenging cases, e.g., tracking partially overlapped objects and suddenly appeared objects. Overall, the granulated system shows a balanced trade-off between speed and accuracy as compared to pixel level learning in tracking and recognition. The concept of using Z-numbers, in providing a granulated linguistic description of a scene, is unique. This gives a more natural interpretation of object recognition in terms of certainty toward scene understanding.