Discrimination of Spectrally Blended Natural Images: Optimisation of the Human Visual System for Encoding Natural Images

AbstractAn important heuristic in developing image processing technologies is to mimic the computational strategies used by humans. Relevant to this, recent studies have shown that the human brain’s processing strategy is closely matched to the characteristics of natural scenes, both in terms of global and local image statistics. However, structural MRI images and natural scenes have fundamental differences: the former are two-dimensional sections through a volume, the latter are projections. MRI image formation is also radically different from natural image formation, involving acquisition in Fourier space, followed by several filtering and processing steps that all have the potential to alter image statistics. As a consequence, aspects of the human visual system that are finely-tuned to processing natural scenes may not be equally well-suited for MRI images, and identification of the differences between MRI images and natural scenes may lead to improved machine analysis of MRI.With these considerations in mind, we analyzed spectra and local image statistics of MRI images in several databases including T1 and FLAIR sequence types and of simulated MRI images,[1]–[6] and compared this analysis to a parallel analysis of natural images[7] and visual sensitivity[7][8]. We found substantial differences between the statistical features of MRI images and natural images. Power spectra of MRI images had a steeper slope than that of natural images, indicating a lack of scale invariance. Independent of this, local image statistics of MRI and natural images differed: compared to natural images, MRI images had smaller variations in their local two-point statistics and larger variations in their local three-point statistics – to which the human visual system is relatively insensitive. Our findings were consistent across MRI databases and simulated MRI images, suggesting that they result from brain geometry at the scale of MRI resolution, rather than characteristics of specific imaging and reconstruction methods.

Download Full-text

Object Recognition Inspiring HVS

Indonesian Journal of Electrical Engineering and Computer Science ◽

10.11591/ijeecs.v12.i2.pp783-793 ◽

2018 ◽

Vol 12 (2) ◽

pp. 783

Author(s):

Mohammadesmaeil Akbarpour ◽

Nasser Mehrshad ◽

Seyyed-Mohammad Razavi

Keyword(s):

Object Recognition ◽

Visual System ◽

Computational Modeling ◽

Hierarchical Model ◽

Human Visual System ◽

Visual Pathway ◽

Natural Images ◽

Significant Performance

<p><span>Human recognize objects in complex natural images very fast within a fraction of a second. Many computational object recognition models inspired from this powerful ability of human. The Human Visual System (HVS) recognizes object in several processing layers which we know them as hierarchically model. Due to amazing complexity of HVS and the connections in visual pathway, computational modeling of HVS directly from its physiology is not possible. So it considered as a some blocks and each block modeled separately. One models inspiring of HVS is HMAX which its main problem is selecting patches in random way. As HMAX is a hierarchical model, HMAX can enhanced with enhancing each layer separately. In this paper instead of random patch extraction, Desirable Patches for HMAX (DPHMAX) will extracted. HVS for extracting patch first selected patches with more information. For simulating this block patches with more variance will be selected. Then HVS will chose patches with more similarity in a class. For simulating this block one algorithm is used. For evaluating proposed method, Caltech 5 and Caltech101 datasets are used. Results show that the proposed method (DPMAX) provides a significant performance over HMAX and other models with the same framework.</span></p>

Download Full-text

How Sensitive Is the Human Visual System to the Local Statistics of Natural Images?

PLoS Computational Biology ◽

10.1371/journal.pcbi.1002873 ◽

2013 ◽

Vol 9 (1) ◽

pp. e1002873 ◽

Cited By ~ 30

Author(s):

Holly E. Gerhard ◽

Felix A. Wichmann ◽

Matthias Bethge

Keyword(s):

Visual System ◽

Human Visual System ◽

Natural Images ◽

Local Statistics

Download Full-text

HCNN: A Neural Network Model for Combining Local and Global Features Towards Human-Like Classification

International Journal of Pattern Recognition and Artificial Intelligence ◽

10.1142/s0218001416550041 ◽

2015 ◽

Vol 30 (01) ◽

pp. 1655004 ◽

Cited By ~ 3

Author(s):

Tielin Zhang ◽

Yi Zeng ◽

Bo Xu

Keyword(s):

Neural Network ◽

Visual System ◽

Image Classification ◽

Network Model ◽

Neural Network Model ◽

Human Visual System ◽

Experimental Results ◽

Natural Images ◽

Global Features ◽

Local And Global Features

Brain-inspired algorithms such as convolutional neural network (CNN) have helped machine vision systems to achieve state-of-the-art performance for various tasks (e.g. image classification). However, CNNs mainly rely on local features (e.g. hierarchical features of points and angles from images), while important global structured features such as contour features are lost. Global understanding of natural objects is considered to be essential characteristics that the human visual system follows, and for developing human-like visual systems, the lost of consideration from this perspective may lead to inevitable failure on certain tasks. Experimental results have proved that well-trained CNN classifier cannot correctly distinguish fooling images (in which some local features from the natural images are chaotically distributed) from natural images. For example, a picture that is composed of yellow–black bars will be recognized as school bus with very high confidence by CNN. On the contrary, human visual system focuses on both the texture and contour features to form representation of images and would not mis-take them. In order to solve the upper problem, we propose a neural network model, named as histogram of oriented gradient (HOG) improved CNN (HCNN), that combines local and global features towards human-like classification based on CNN and HOG. The experimental results on MNIST datasets and part of ImageNet datasets show that HCNN outperforms traditional CNN for object classification with fooling images, which indicates the feasibility, accuracy and potential effectiveness of HCNN for solving image classification problem.

Download Full-text