Learning Nonlinear Visual Processing from Natural Images

The paradigm of computational vision hypothesizes that any visual function -- such as the recognition of your grandparent -- can be replicated by computational processing of the visual input. What are these computations that the brain performs? What should or could they be? Working on the latter question, this dissertation takes the statistical approach, where the suitable computations are attempted to be learned from the natural visual data itself. In particular, we empirically study the computational processing that emerges from the statistical properties of the visual world and the constraints and objectives specified for the learning process. This thesis consists of an introduction and 7 peer-reviewed publications, where the purpose of the introduction is to illustrate the area of study to a reader who is not familiar with computational vision research. In the scope of the introduction, we will briefly overview the primary challenges to visual processing, as well as recall some of the current opinions on visual processing in the early visual systems of animals. Next, we describe the methodology we have used in our research, and discuss the presented results. We have included some additional remarks, speculations and conclusions to this discussion that were not featured in the original publications. We present the following results in the publications of this thesis. First, we empirically demonstrate that luminance and contrast are strongly dependent in natural images, contradicting previous theories suggesting that luminance and contrast were processed separately in natural systems due to their independence in the visual data. Second, we show that simple cell -like receptive fields of the primary visual cortex can be learned in the nonlinear contrast domain by maximization of independence. Further, we provide first-time reports of the emergence of conjunctive (corner-detecting) and subtractive (opponent orientation) processing due to nonlinear projection pursuit with simple objective functions related to sparseness and response energy optimization. Then, we show that attempting to extract independent components of nonlinear histogram statistics of a biologically plausible representation leads to projection directions that appear to differentiate between visual contexts. Such processing might be applicable for priming, \ie the selection and tuning of later visual processing. We continue by showing that a different kind of thresholded low-frequency priming can be learned and used to make object detection faster with little loss in accuracy. Finally, we show that in a computational object detection setting, nonlinearly gain-controlled visual features of medium complexity can be acquired sequentially as images are encountered and discarded. We present two online algorithms to perform this feature selection, and propose the idea that for artificial systems, some processing mechanisms could be selectable from the environment without optimizing the mechanisms themselves. In summary, this thesis explores learning visual processing on several levels. The learning can be understood as interplay of input data, model structures, learning objectives, and estimation algorithms. The presented work adds to the growing body of evidence showing that statistical methods can be used to acquire intuitively meaningful visual processing mechanisms. The work also presents some predictions and ideas regarding biological visual processing.

Maintaining plant species richness by cattle grazing - mesic semi-natural grasslands as focal habitats

Semi-natural grasslands are the most important agricultural areas for biodiversity. The present study investigates the effects of traditional livestock grazing and mowing on plant species richness, the main emphasis being on cattle grazing in mesic semi-natural grasslands. The two reviews provide a thorough assessment of the multifaceted impacts and importance of grazing and mowing management to plant species richness. It is emphasized that livestock grazing and mowing have partially compensated the suppression of major natural disturbances by humans and mitigated the negative effects of eutrophication. This hypothesis has important consequences for nature conservation: A large proportion of European species originally adapted to natural disturbances may be at present dependent on livestock grazing and / or mowing. Furthermore, grazing and mowing are key management methods to mitigate effects of nutrient-enrichment. The species composition and richness in old (continuously grazed), new (grazing restarting 3-8 years ago) and abandoned (over 10 years) pastures differed consistently across a range of spatial scales, and was intermediate in new pastures compared to old and abandoned pastures. In mesic grasslands most plant species were shown to benefit from cattle grazing. Indicator species of biologically valuable grasslands and rare species were more abundant in grazed than in abandoned grasslands. Steep S-SW-facing slopes are the most suitable sites for many grassland plants and should be prioritized in grassland restoration. The proportion of species trait groups benefiting from grazing was higher in mesic semi-natural grasslands than in dry and wet grasslands. Consequently, species trait responses to grazing and the effectiveness of the natural factors limiting plant growth may be intimately linked High plant species richness of traditionally mowed and grazed areas is explained by numerous factors which operate on different spatial scales. Particularly important for maintaining large scale plant species richness are evolutionary and mitigation factors. Grazing and mowing cause a shift towards the conditions that have occurred during the evolutionary history of European plant species by modifying key ecological factors (nutrients, pH and light). The results of this Dissertation suggest that restoration of semi-natural grasslands by private farmers is potentially a useful method to manage biodiversity in the agricultural landscape. However, the quality of management is commonly improper, particularly due to financial constraints. For enhanced success of restoration, management regulations in the agri-environment scheme need to be defined more explicitly and the scheme should be revised to encourage management of biodiversity.

Local systems, global impacts : Using life cycle assessment to analyse the potential and constraints of industrial symbioses

Human activities extract and displace different substances and materials from the earth s crust, thus causing various environmental problems, such as climate change, acidification and eutrophication. As problems have become more complicated, more holistic measures that consider the origins and sources of pollutants have been called for. Industrial ecology is a field of science that forms a comprehensive framework for studying the interactions between the modern technological society and the environment. Industrial ecology considers humans and their technologies to be part of the natural environment, not separate from it. Industrial operations form natural systems that must also function as such within the constraints set by the biosphere. Industrial symbiosis (IS) is a central concept of industrial ecology. Industrial symbiosis studies look at the physical flows of materials and energy in local industrial systems. In an ideal IS, waste material and energy are exchanged by the actors of the system, thereby reducing the consumption of virgin material and energy inputs and the generation of waste and emissions. Companies are seen as part of the chains of suppliers and consumers that resemble those of natural ecosystems. The aim of this study was to analyse the environmental performance of an industrial symbiosis based on pulp and paper production, taking into account life cycle impacts as well. Life Cycle Assessment (LCA) is a tool for quantitatively and systematically evaluating the environmental aspects of a product, technology or service throughout its whole life cycle. Moreover, the Natural Step Sustainability Principles formed a conceptual framework for assessing the environmental performance of the case study symbiosis (Paper I). The environmental performance of the case study symbiosis was compared to four counterfactual reference scenarios in which the actors of the symbiosis operated on their own. The research methods used were process-based life cycle assessment (LCA) (Papers II and III) and hybrid LCA, which combines both process and input-output LCA (Paper IV). The results showed that the environmental impacts caused by the extraction and processing of the materials and the energy used by the symbiosis were considerable. If only the direct emissions and resource use of the symbiosis had been considered, less than half of the total environmental impacts of the system would have been taken into account. When the results were compared with the counterfactual reference scenarios, the net environmental impacts of the symbiosis were smaller than those of the reference scenarios. The reduction in environmental impacts was mainly due to changes in the way energy was produced. However, the results are sensitive to the way the reference scenarios are defined. LCA is a useful tool for assessing the overall environmental performance of industrial symbioses. It is recommended that in addition to the direct effects, the upstream impacts should be taken into account as well when assessing the environmental performance of industrial symbioses. Industrial symbiosis should be seen as part of the process of improving the environmental performance of a system. In some cases, it may be more efficient, from an environmental point of view, to focus on supply chain management instead.