Atjaunināt sīkdatņu piekrišanu

E-grāmata: Understanding Vision: Theory, Models, and Data [Oxford Scholarship Online E-books]

(Professor, Department of Computer Science, University College London, UK)
  • Formāts: 396 pages
  • Izdošanas datums: 22-May-2014
  • Izdevniecība: Oxford University Press
  • ISBN-13: 9780199564668
Citas grāmatas par šo tēmu:
  • Oxford Scholarship Online E-books
  • Cena pašlaik nav zināma
Understanding Vision: Theory, Models, and DataPalielināt
  • Formāts: 396 pages
  • Izdošanas datums: 22-May-2014
  • Izdevniecība: Oxford University Press
  • ISBN-13: 9780199564668
Citas grāmatas par šo tēmu:
Wiley OnlineMore info about Oxford Scholarship Online e-books
Rokasgrāmatas mājas lapa: http://www.oxfordscholarship.com/view/10.1093/acprof:oso/9780199564668.001.0001/acprof-9780199564668
While the field of vision science has grown significantly in the past two decades, there have been few comprehensive books that showed readers how to adopt a computional approach to understanding visual perception, along with the underlying mechanisms in the brain.

Understanding Vision explains the computational principles and models of biological visual processing, and in particular, of primate vision. The book is written in such a way that vision scientists, unfamiliar with mathematical details, should be able to conceptually follow the theoretical principles and their relationship with physiological, anatomical, and psychological observations, without going through the more mathematical pages. For those with a physical science background, especially those from machine vision, this book serves as an analytical introduction to biological vision. It can be used as a textbook or a reference book in a vision course, or a computational neuroscience course for graduate students or advanced undergraduate students. It is also suitable for self-learning by motivated readers.

in addition, for those with a focused interest in just one of the topics in the book, it is feasible to read just the chapter on this topic without having read or fully comprehended the other chapters. In particular, Chapter 2 presents a brief overview of experimental observations on biological vision; Chapter 3 is on encoding of visual inputs, Chapter 5 is on visual attentional selection driven by sensory inputs, and Chapter 6 is on visual perception or decoding.

Including many examples that clearly illustrate the application of computational principles to experimental observations,Understanding Vision is valuable for students and researchers in computational neuroscience, vision science, machine and computer vision, as well as physicists interested in visual processes.
1 Approach and scope 1(15)
1.1 The approach
1(3)
1.1.1 Data, models, and theory
1(2)
1.1.2 From physiology to behavior and back via theory and models
3(1)
1.2 The problem of vision
4(12)
1.2.1 Visual tasks and subtasks
5(2)
1.2.2 Vision seen through visual encoding, selection, and decoding
7(3)
1.2.3 Visual encoding in retina and V1
10(2)
1.2.4 Visual selection and V1's role in it
12(2)
1.2.5 Visual decoding and its associated brain areas
14(2)
2 A very brief introduction of what is known about vision experimentally 16(51)
2.1 Neurons, neural circuits, and brain regions
16(6)
2.1.1 Neurons, somas, dendrites, axons, and action potentials
16(1)
2.1.2 A simple neuron model
17(1)
2.1.3 Random processes of action potential generation in neurons
18(1)
2.1.4 Synaptic connections, neural circuits, and brain areas
18(1)
2.1.5 Visual processing areas along the visual pathway
19(3)
2.2 Retina
22(17)
2.2.1 Receptive fields of retinal ganglion cells
22(3)
2.2.2 Sensitivity to sinusoidal gratings, and contrast sensitivity curves
25(5)
2.2.3 Responses to spatiotemporal inputs
30(4)
2.2.4 P and M cells
34(1)
2.2.5 Color processing in the retina
35(2)
2.2.6 Spatial sampling in the retina
37(1)
2.2.7 LGN on the pathway from the retina to V1
38(1)
2.3 V1
39(15)
2.3.1 The retinotopic map
39(1)
2.3.2 The receptive fields in V1-the feature detectors
40(1)
2.3.3 Orientation selectivity, bar and edge detectors
41(1)
2.3.4 Spatial frequency tuning and multiscale coding
42(1)
2.3.5 Temporal and motion direction selectivity
43(3)
2.3.6 Ocular dominance and disparity selectivity
46(2)
2.3.7 Color selectivity of V1 neurons
48(1)
2.3.8 Complex cells
48(4)
2.3.9 The influences on a V1 neuron's response from contextual stimuli outside the receptive field
52(2)
2.4 Higher visual areas
54(6)
2.4.1 Two processing streams
54(1)
2.4.2 V2
55(2)
2.4.3 MT (V5)
57(2)
2.4.4 V4
59(1)
2.4.5 IT and temporal cortical areas for object recognition
60(1)
2.5 Eye movements, their associated brain regions, and links with attention
60(3)
2.5.1 Close link between eye movements and attention
62(1)
2.6 Top-down attention and neural responses
63(2)
2.7 Behavioral studies on vision
65(2)
3 The efficient coding principle 67(110)
3.1 A brief introduction to information theory
68(9)
3.1.1 Measuring information
68(2)
3.1.2 Information transmission, information channels, and mutual information
70(4)
3.1.3 Information redundancy, representation efficiency, and error correction
74(3)
3.2 Formulation of the efficient coding principle
77(6)
3.2.1 An optimization problem
77(2)
3.2.2 Exposition
79(4)
3.3 Efficient neural sampling in the retina
83(7)
3.3.1 Contrast sampling in a fly's compound eye
83(2)
3.3.2 Spatial sampling by receptor distribution on the retina
85(4)
3.3.3 Optimal color sampling by the cones
89(1)
3.4 Efficient coding by visual receptive field transforms
90(6)
3.4.1 The general analytical solution for efficient coding of Gaussian signals
91(5)
3.5 Case study: stereo coding in V1 as an efficient transform of inputs in the dimension of ocularity
96(24)
3.5.1 Principal component analysis K0
98(4)
3.5.2 Gain control
102(3)
3.5.3 Contrast enhancement, decorrelation, and whitening in the high S/N regime
105(1)
3.5.4 Many equivalent solutions of optimal encoding
106(2)
3.5.5 Smoothing and output correlation in the low S/N region
108(2)
3.5.6 A special, most local, class of optimal coding
110(1)
3.5.7 Adaptation of the optimal code to the statistics of the input environment
110(7)
3.5.8 A psychophysical test of the adaptation of the efficient stereo coding
117(3)
3.5.9 How might one test the predictions physiologically?
120(1)
3.6 The efficient receptive field transforms in space, color, time, and scale in the retina and V1
120(50)
3.6.1 Efficient spatial coding in the retina
123(11)
3.6.2 Efficient coding in time
134(4)
3.6.3 Efficient coding in color
138(4)
3.6.4 Coupling space and color coding in the retina
142(5)
3.6.5 Spatial coding in V1
147(7)
3.6.6 Coupling the spatial and color coding in V1
154(7)
3.6.7 Coupling spatial coding with stereo coding in V1-coding disparity
161(3)
3.6.8 Coupling space and time coding in the retina and V1
164(3)
3.6.9 V1 neurons tuned simultaneously to multiple feature dimensions
167(3)
3.7 The efficient code, and the related sparse code, in low noise limit by numerical simulations
170(3)
3.7.1 Sparse coding
171(2)
3.8 How to get efficient codes by developmental rules and unsupervised learning
173(4)
3.8.1 Learning for a single encoding neuron
174(1)
3.8.2 Learning simultaneously for multiple encoding neurons
175(2)
4 V1 and information coding 177(12)
4.1 Pursuit of efficient coding in V1 by reducing higher order redundancy
177(9)
4.1.1 Higher order statistics contain much of the meaningful information about visual objects
178(2)
4.1.2 Characterizing higher order statistics
180(3)
4.1.3 Efforts to understand V1 neural properties from the perspective of reducing higher order redundancy
183(2)
4.1.4 Higher order redundancy in natural images is only a very small fraction of the total redundancy
185(1)
4.2 Problems in understanding V1 solely based on efficient coding
186(1)
4.3 Multiscale and overcomplete representation in V1 is useful for invariant object recognition from responses of selected neural subpopulations
187(2)
4.3.1 Information selection, amount, and meaning
188(1)
5 The V1 hypothesis-creating a bottom-up saliency map for preattentive selection and segmentation 189(126)
5.1 Visual selection and visual saliency
189(12)
5.1.1 Visual selection-top-down and bottom-up selections
189(6)
5.1.2 A brief overview of visual search and segmentation-behavioral studies of saliency
195(2)
5.1.3 Saliency regardless of visual input features
197(3)
5.1.4 A quick review of what we should expect about saliencies and a saliency map
200(1)
5.2 The V1 saliency hypothesis
201(8)
5.2.1 Detailed formulation of the V1 saliency hypothesis
202(2)
5.2.2 Intracortical interactions in V1 as mechanisms to compute saliency
204(2)
5.2.3 Reading out the saliency map
206(1)
5.2.4 Statistical and operational definitions of saliency
207(1)
5.2.5 Overcomplete representation in V1 for the role of saliency
208(1)
5.3 A hallmark of the saliency map in V1-attention capture by an ocular singleton which is barely distinctive to perception
209(6)
5.3.1 Food for thought: looking (acting) before or without seeing
215(1)
5.4 Testing and understanding the V1 saliency map in a V1 model
215(37)
5.4.1 The V1 model: its neural elements, connections, and desired behavior
216(6)
5.4.2 Calibration of the V1 model to biological reality
222(3)
5.4.3 Computational requirements on the dynamic behavior of the model
225(2)
5.4.4 Applying the V1 model to visual search and visual segmentation
227(20)
5.4.5 Other effects of the saliency mechanisms-figure-ground segmentation and the medial axis effect
247(3)
5.4.6 Input contrast dependence of the contextual influences
250(1)
5.4.7 Reflections from the V1 model
250(2)
5.5 Additional psychophysical tests of the V1 saliency hypothesis
252(17)
5.5.1 The feature-blind "auction"-maximum rather than summation over features
252(5)
5.5.2 The fingerprints of colinear facilitation in V1
257(3)
5.5.3 The fingerprint of V1's conjunctive cells
260(6)
5.5.4 A zero-parameter quantitative prediction and its experimental test
266(3)
5.5.5 Reflections-from behavior back to physiology via the V1 saliency hypothesis
269(1)
5.6 The roles of V1 and other cortical areas in visual selection
269(10)
5.6.1 Using visual depth feature to probe contributions of extrastriate cortex to atten- tional control
271(4)
5.6.2 Salient but indistinguishable inputs activate early visual cortical areas but not the parietal and frontal areas
275(4)
5.7 V1's role beyond saliency-selection versus decoding, periphery versus central vision
279(6)
5.7.1 Implications for the functional roles of visual cortical areas based on their repre- sentations of the visual field
281(1)
5.7.2 Saliency, visual segmentation, and visual recognition
282(3)
5.8 Nonlinear V1 neural dynamics for saliency and preattentive segmentation
285(28)
5.8.1 A minimal model of the primary visual cortex for saliency computation
286(13)
5.8.2 Dynamic analysis of the V1 model and constraints on the neural connections
299(13)
5.8.3 Extensions and generalizations
312(1)
5.9 Appendix: parameters in the V1 model
313(2)
6 Visual recognition as decoding 315(49)
6.1 Definition of visual decoding
315(2)
6.2 Some notable observations about visual recognition
317(9)
6.2.1 Recognition is after an initial selection or segmentation
317(1)
6.2.2 Object invariance
318(1)
6.2.3 Is decoding the shape of an object in the attentional spotlight a default routine?
319(2)
6.2.4 Recognition by imagination or input synthesis
321(2)
6.2.5 Visual perception can be ambiguous or unambiguous
323(2)
6.2.6 Neural substrates for visual decoding
325(1)
6.3 Visual decoding from neural responses
326(21)
6.3.1 Example: decoding motion direction from MT neural responses
327(3)
6.3.2 Example: discriminating two inputs based on photoreceptor responses
330(2)
6.3.3 Example: discrimination by decoding the V1 neural responses
332(2)
6.3.4 Example: light wavelength discrimination by decoding from cone responses
334(4)
6.3.5 Perception, including illusion, of a visual feature value by neural population de- coding
338(7)
6.3.6 Poisson-like neural noise and increasing perceptual performance for stronger visual inputs
345(1)
6.3.7 Low efficiency of sensory information utilization by the central visual system
345(1)
6.3.8 Transduction and central inefficiencies in the framework of encoding, attentional selection, and decoding
346(1)
6.4 Bayesian inference and the influence of prior belief in visual decoding
347(14)
6.4.1 The Bayesian framework
348(1)
6.4.2 Bayesian visual inference is highly complex unless the number and the dimen sions of possible percepts are restricted
349(1)
6.4.3 Behavioral evidence for Bayesian visual inference
350(11)
6.5 The initial visual recognition, feedforward mechanisms, and recurrent neural connections
361(3)
6.5.1 The fast speed of coarse initial recognition by the primate visual system
361(1)
6.5.2 Object detection and recognition by models of hierarchical feedforward networks
361(2)
6.5.3 Combining feedforward and feedback intercortical mechanisms, and recurrent intracortical mechanisms, for object inference
363(1)
7 Epilogue 364(3)
7.1 Our ignorance of vision viewed from the perspective of vision as encoding, selection, and decoding
364(1)
7.2 Computational vision
365(2)
References 367(13)
Index 380
Li Zhaoping obtained her Ph.D. in physics in 1989 from the California Institute of Technology. In 1998, she helped to found the Gatsby Computational Neuroscience Unit in University College London, where she is currently a professor in Computational Neuroscience in its computer science department.
Permanent link: https://www.kriso.lv/db/9780199564668_pe.html