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.

MCBoost: Multiple classifier boosting for perceptual co-clustering of images and visual features

We present a new co-clustering problem of images and visual features. The problem involves a set of non-object images in addition to a set of object images and features to be co-clustered. Co-clustering is performed in a way that maximises discrimination of object images from non-object images, thus emphasizing discriminative features. This provides a way of obtaining perceptual joint-clusters of object images and features. We tackle the problem by simultaneously boosting multiple strong classifiers which compete for images by their expertise. Each boosting classifier is an aggregation of weak-learners, i.e. simple visual features. The obtained classifiers are useful for object detection tasks which exhibit multimodalities, e.g. multi-category and multi-view object detection tasks. Experiments on a set of pedestrian images and a face data set demonstrate that the method yields intuitive image clusters with associated features and is much superior to conventional boosting classifiers in object detection tasks.

Study of the images for the Visual Arts Education in Secondary Education textbooks: educational implications

In this paper we present the first data from the research conducted to determine the relationship between traditional visual arts and other forms of visual culture closer to the experiences of high school youth. The hypothesis of this research is that while students are nurtured and live primarily with the images provided by the media culture, their textbooks basically refer to the more traditional art images. The research has been limited to a review and analysis of the most common educational materials for teaching visual arts in high school. After the systematization and analysis of the images appeared in textbooks, we have detected three major types: the artistics, those who belong to media culture and others. The most relevant conclusions indicate that: there are hardly any connections between different types of images, they offer a very traditional view of art and they are far removed from the experiences of young book users.

Emotional response to images of wind turbines: A psychophysiological study of their visual impact on the landscape

Social acceptance for wind turbines is variable, providing a challenge to the implementation of this energy source. Psychological research could contribute to the science of climate change. Here we focus on the emotional responses to the visual impact of wind turbines on the landscape, a factor which dominates attitudes towards this technology. Participants in the laboratory viewed images of turbines and other constructions (churches, pylons and power-plants) against rural scenes, and provided psychophysiological and self-report measures of their emotional reactions. We hypothesised that the emotional response to wind turbines would be more negative and intense than to control objects, and that this difference would be accentuated for turbine opponents. As predicted, the psychophysiological response to turbines was stronger than the response to churches, but did not differ from that of other industrial constructions. In contrast with predictions, turbines were rated as less aversive and more calming compared with other industrial constructions, and equivalent to churches. Supporters and non-supporters did not differ significantly from each other. We discuss how a methodology using photo manipulations and emotional self-assessments can help estimate the emotional reaction to the visual impact on the landscape at the planning stage for new wind turbine applications.

Beyond the analogon; the "transfer of the referent" : the relationship between the external world experienced through culturally generated perceptual models constructed in the human mind, and the transfer of these images, via visual language, onto referen

Within 20th. Century art, the concept of the ‘normative’ image, as an attribute of things, has been challenged. As a consequence, paintings must now picture the ‘real’ world in other ways, incorporating knowledge and meaning beyond the analogon. Such descriptive representations were revealed as paradigmatic, rather than incontrovertible fact. Dependent on pre-conceived notions of stereotypicality, these descriptive images relied on surface illumination. My thesis explores images of things in the world as culturally inspired and information based. I examine paintings and sculptures of other cultures, such as black African, and other historical periods such as the Medieval, which reveal metonymously the basis for variations in representations of the ‘real’ world. The new enhanced representations which Modern artists created in their work were denigrated as deviant from the absolute ‘normative’ or regarded as distortions for purely mannerist and stylistic reasons. Postmodern research has reassessed them as multiple or extended imagings in whose facture new knowledge and human responses can be incorporated. These new forms of representation can be regarded as theoretical constructs rather than stylised depictions of appearance. In this way referents are transferred through the mind onto objects and vistas in the real world to align with our developed view of the physical world and better our understanding of humanity’s symbiotic relationship with nature.

Visual narratives in the children's picture story book : an investigation of the extent to which the formal structuring of a visual narrative retrieves a pastiche of past experiences and images which in turn shapes and redirects the story

Throughout this research, the notion that illustrators of children's books embark on two types of activity has been reinforced at every turn. On one hand the artist acknowledges the external world by organising images of actions and events in the contexts of place and time. This process involves bringing ideas into a physical form and demands the structuring of characters, settings, and story development. Planning and decisions are informed by imperatives that recognise the need for conventions of articulation and communication to a particular target audience. These then become a mayor priority of bookmaking and are constantly impacted on by publishers1 demands and ethical constraints. The other perspective sees the illustrator as expresser where the core of visual narratives for children celebrates the potency of imagination. Here dreams, fantasies, memories and the unconscious become the conduits to shaping sequential images. The artist is engaged not simply in visually telling a story, but rather telling facets of his or her own story. This exegesis traces the evolution of my own picture story book Eddie's Fantastic Fortnight published by Five Mile Press Publishers in tandem with the insights and reflections of five of Australia's most prominent illustrators. It examines whether the structure invested in a visual narrative liberates expressive response, ascribing to the premise that bookmaking plays an informing role to imagination. Equally it adopts the alternative position which asserts that the essence of children's books is indeed fantasy, memory and dreams. This proposition views imagination and inspiration as the primary catalyst around which illustrators build their narrative. In the often lengthy processes of bookmaking, these considerations constantly shift. I have attempted to explore and reveal these mobile and ever changing priorities, not only in my own work, but also through leading exponents in the field.