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.

Deus vult: Images of crusader violence c. 1095-1100.

This study sets out to provide new information about the interaction between abstract religious ideas and actual acts of violence in the early crusading movement. The sources are asked, whether such a concept as religious violence can be sorted out as an independent or distinguishable source of aggression at the moment of actual bloodshed. The analysis concentrates on the practitioners of sacred violence, crusaders and their mental processing of the use of violence, the concept of the violent act, and the set of values and attitudes defining this concept. The scope of the study, the early crusade movement, covers the period from late 1080 s to the crusader conquest of Jerusalem in 15 July 1099. The research has been carried out by contextual reading of relevant sources. Eyewitness reports will be compared with texts that were produced by ecclesiastics in Europe. Critical reading of the texts reveals both connecting ideas and interesting differences between them. The sources share a positive attitude towards crusading, and have principally been written to propagate the crusade institution and find new recruits. The emphasis of the study is on the interpretation of images: the sources are not asked what really happened in chronological order, but what the crusader understanding of the reality was like. Fictional material can be even more crucial for the understanding of the crusading mentality. Crusader sources from around the turn of the twelfth century accept violent encounters with non-Christians on the grounds of external hostility directed towards the Christian community. The enemies of Christendom can be identified with either non-Christians living outside the Christian society (Muslims), non-Christians living within the Christian society (Jews) or Christian heretics. Western Christians are described as both victims and avengers of the surrounding forces of diabolical evil. Although the ideal of universal Christianity and gradual eradication of the non-Christian is present, the practical means of achieving a united Christendom are not discussed. The objective of crusader violence was thus entirely Christian: the punishment of the wicked and the restoration of Christian morals and the divine order. Meanwhile, the means used to achieve these objectives were not. Given the scarcity of written regulations concerning the use of force in bello, perceptions concerning the practical use of violence were drawn from a multitude of notions comprising an adaptable network of secular and ecclesiastical, pre-Christian and Christian traditions. Though essentially ideological and often religious in character, the early crusader concept of the practise of violence was not exclusively rooted in Christian thought. The main conclusion of the study is that there existed a definable crusader ideology of the use of force by 1100. The crusader image of violence involved several levels of thought. Predominantly, violence indicates a means of achieving higher spiritual rewards; eternal salvation and immortal glory.

Using Mathematical Morphology for Geometric Music Retrieval

The usual task in music information retrieval (MIR) is to find occurrences of a monophonic query pattern within a music database, which can contain both monophonic and polyphonic content. The so-called query-by-humming systems are a famous instance of content-based MIR. In such a system, the user's hummed query is converted into symbolic form to perform search operations in a similarly encoded database. The symbolic representation (e.g., textual, MIDI or vector data) is typically a quantized and simplified version of the sampled audio data, yielding to faster search algorithms and space requirements that can be met in real-life situations. In this thesis, we investigate geometric approaches to MIR. We first study some musicological properties often needed in MIR algorithms, and then give a literature review on traditional (e.g., string-matching-based) MIR algorithms and novel techniques based on geometry. We also introduce some concepts from digital image processing, namely the mathematical morphology, which we will use to develop and implement four algorithms for geometric music retrieval. The symbolic representation in the case of our algorithms is a binary 2-D image. We use various morphological pre- and post-processing operations on the query and the database images to perform template matching / pattern recognition for the images. The algorithms are basically extensions to classic image correlation and hit-or-miss transformation techniques used widely in template matching applications. They aim to be a future extension to the retrieval engine of C-BRAHMS, which is a research project of the Department of Computer Science at University of Helsinki.

Digital Photography in the Diagnosis and Follow-up of Ocular Diseases

Purpose: The aim of the present study was to develop and test new digital imaging equipment and methods for diagnosis and follow-up of ocular diseases. Methods: The whole material comprised 398 subjects (469 examined eyes), including 241 patients with melanocytic choroidal tumours, 56 patients with melanocytic iris tumours, 42 patients with diabetes, a 52-year old patient with chronic phase of VKH disease, a 30-year old patient with an old blunt eye injury, and 57 normal healthy subjects. Digital 50° (Topcon TRC 50 IA) and 45° (Canon CR6-45NM) fundus cameras, a new handheld digital colour videocamera for eye examinations (MediTell), a new subtraction method using the Topcon Image Net Program (Topcon corporation, Tokyo, Japan), a new method for digital IRT imaging of the iris we developed, and Zeiss photoslitlamp with a digital camera body were used for digital imaging. Results: Digital 50° red-free imaging had a sensitivity of 97.7% and two-field 45° and 50° colour imaging a sensitivity of 88.9-94%. The specificity of the digital 45°-50° imaging modalities was 98.9-100% versus the reference standard and ungradeable images that were 1.2-1.6%. By using the handheld digital colour video camera only, the optic disc and central fundus located inside 20° from the fovea could be recorded with a sensitivity of 6.9% for detection of at least mild NPDR when compared with the reference standard. Comparative use of digital colour, red-free, and red light imaging showed 85.7% sensitivity, 99% specificity, and 98.2 % exact agreement versus the reference standard in differentiation of small choroidal melanoma from pseudomelanoma. The new subtraction method showed growth in four of 94 melanocytic tumours (4.3%) during a mean ±SD follow-up of 23 ± 11 months. The new digital IRT imaging of the iris showed the sphincter muscle and radial contraction folds of Schwalbe in the pupillary zone and radial structural folds of Schwalbe and circular contraction furrows in the ciliary zone of the iris. The 52-year-old patient with a chronic phase of VKH disease showed extensive atrophy and occasional pigment clumps in the iris stroma, detachment of the ciliary body with severe ocular hypotony, and shallow retinal detachment of the posterior pole in both eyes. Infrared transillumination imaging and fluorescein angiographic findings of the iris showed that IR translucence (p=0.53), complete masking of fluorescence (p=0.69), presence of disorganized vessels (p=0.32), and fluorescein leakage (p=1.0) at the site of the lesion did not differentiate an iris nevus from a melanoma. Conclusions: Digital 50° red-free and two-field 50° or 45° colour imaging were suitable for DR screening, whereas the handheld digital video camera did not fulfill the needs of DR screening. Comparative use of digital colour, red-free and red light imaging was a suitable method in the differentiation of small choroidal melanoma from different pseudomelanomas. The subtraction method may reveal early growth of the melanocytic choroidal tumours. Digital IRT imaging may be used to study changes of the stroma and posterior surface of the iris in various diseases of the uvea. It contributed to the revealment of iris atrophy and serous detachment of the ciliary body with ocular hypotony together with the shallow retinal detachment of the posterior pole as new findings of the chronic phase of VKH disease. Infrared translucence and angiographic findings are useful in differential diagnosis of melanocytic iris tumours, but they cannot be used to determine if the lesion is benign or malignant.