Exploring the role of texture features in deep convolutional neural networks: Insights from Portilla-Simoncelli statistics

It is well-understood that the performance of Deep Convolutional Neural Networks (DCNNs) in image recognition tasks is influenced not only by shape but also by texture information. Despite this, understanding the internal representations of DCNNs remains a challenging task. This study employs a simplified version of the Portilla-Simoncelli Statistics, termed "minPS," to explore how texture information is represented in a pre-trained VGG network. Using minPS features extracted from texture images, we perform a sparse regression on the activations across various channels in VGG layers. Our findings reveal that channels in the early to middle layers of the VGG network can be effectively described by minPS features. Additionally, we observe that the explanatory power of minPS sub-groups evolves as one ascends the network hierarchy. Specifically, sub-groups termed Linear Cross Scale (LCS) and Energy Cross Scale (ECS) exhibit weak explanatory power for VGG channels. To investigate the relationship further, we compare the original texture images with their synthesized counterparts, generated using VGG, in terms of minPS features. Our results indicate that the absence of certain minPS features suggests their non-utilization in VGG's internal representations.

Distorted image classification using neural activation pattern matching loss

In image classification, a deep neural network (DNN) that is trained on undistorted images constitutes an effective decision boundary. Unfortunately, this boundary does not support distorted images, such as noisy or blurry ones, leading to accuracy drop-off. As a simple approach for classifying distorted images as well as undistorted ones, previous methods have optimized the trained DNN again on both kinds of images. However, in these methods, the decision boundary may become overly complicated during optimization because there is no regularization of the decision boundary. Consequently, this decision boundary limits efficient optimization. In this paper, we study a simple yet effective decision boundary for distorted image classification through the use of a novel loss, called a "neural activation pattern matching (NAPM) loss". The NAPM loss is based on recent findings that the decision boundary is a piecewise linear function, where each linear segment is constructed from a neural activation pattern in the DNN when an image is fed to it. The NAPM loss extracts the neural activation patterns when the distorted image and its undistorted version are fed to the DNN and then matches them with each other via the sigmoid cross-entropy. Therefore, it constrains the DNN to classify the distorted image and its undistorted version by the same linear segment. As a result, our loss accelerates efficient optimization by preventing the decision boundary from becoming overly complicated. Our experiments demonstrate that our loss increases the accuracy of the previous methods in all conditions evaluated.

Development of spectral decomposition based on Bayesian information criterion with estimation of confidence interval

We develop an automatic peak fitting algorithm using the Bayesian information criterion (BIC) fitting method with confidence-interval estimation in spectral decomposition. First, spectral decomposition is carried out by adopting the Bayesian exchange Monte Carlo method for various artificial spectral data, and the confidence interval of fitting parameters is evaluated. From the results, an approximated model formula that expresses the confidence interval of parameters and the relationship between the peak-to-peak distance and the signal-to-noise ratio is derived. Next, for real spectral data, we compare the confidence interval of each peak parameter obtained using the Bayesian exchange Monte Carlo method with the confidence interval obtained from the BIC-fitting with the model selection function and the proposed approximated formula. We thus confirm that the parameter confidence intervals obtained using the two methods agree well. It is therefore possible to not only simply estimate the appropriate number of peaks by BIC-fitting but also obtain the confidence interval of fitting parameters.

Simultaneous Estimation of Nongaussian Components and Their Correlation Structure

The statistical dependencies that independent component analysis (ICA) cannot remove often provide rich information beyond the linear independent components. It would thus be very useful to estimate the dependency structure from data. While such models have been proposed, they have usually concentrated on higher-order correlations such as energy (square) correlations. Yet linear correlations are a fundamental and informative form of dependency in many real data sets. Linear correlations are usually completely removed by ICA and related methods so they can only be analyzed by developing new methods that explicitly allow for linearly correlated components. In this article, we propose a probabilistic model of linear nongaussian components that are allowed to have both linear and energy correlations. The precision matrix of the linear components is assumed to be randomly generated by a higher-order process and explicitly parameterized by a parameter matrix. The estimation of the parameter matrix is shown to be particularly simple because using score-matching (Hyvärinen, 2005 ), the objective function is a quadratic form. Using simulations with artificial data, we demonstrate that the proposed method improves the identifiability of nongaussian components by simultaneously learning their correlation structure. Applications on simulated complex cells with natural image input, as well as spectrograms of natural audio data, show that the method finds new kinds of dependencies between the components.

Formation of a direction map by projection learning using Kohonen's self-organization map

In this paper, we propose a modification of Kohonen's self-organization map (SOM) algorithm. When the input signal space is not convex, some reference vectors of SOM can protrude from it. The input signal space must be convex to keep all the reference vectors fixed on it for any updates. Thus, we introduce a projection learning method that fixes the reference vectors onto the input signal space. This version of SOM can be applied to a non-convex input signal space. We applied SOM with projection learning to a direction map observed in the primary visual cortex of area 17 of ferrets, and area 18 of cats. Neurons in those areas responded selectively to the orientation of edges or line segments, and their directions of motion. Some iso-orientation domains were subdivided into selective regions for the opposite direction of motion. The abstract input signal space of the direction map described in the manner proposed by Obermayer and Blasdel [(1993) J Neurosci 13: 4114-4129] is not convex. We successfully used SOM with projection learning to reproduce a direction-orientation joint map.

Relationship between the placental location and the flow velocity waveforms of bilateral uterine arteries

The relationship between the placental location and the resistance index (RI) of the ascending branch of bilateral uterine arteries was examined with continuous Doppler ultrasonography, in 48 healthy pregnant women in the 37th to 40th week of gestation. When the placental location was central, there was little difference in the RI between the right and left uterine artery (fundus; 0.61 +/- 0.04 versus 0.58 +/- 0.03, NS, body; 0.48 +/- 0.04 versus 0.47 +/- 0.03, NS, and lower segment; 0.28 +/- 0.03 versus 0.30 +/- 0.04, NS). The RI became smaller when the level of placental attachment was lower toward the cervix. In cases where the placental location has shifted to the right or left of the uterine body, the RI on the placental side was smaller than that of the contralateral uterine artery (0.39 +/- 0.04 versus 0.62 +/- 0.03, p less than 0.01). The physiological and clinical significance of these results are discussed in the following report.

Significance of cardiovascular malformations in cystic hygroma: a new interpretation of the pathogenesis

Fetuses with cystic hygroma or loose skin of the neck were studied chromosomally and phenotypically to clarify the relation between neck abnormality and cardiovascular malformations. Of 12 fetuses, 9 had chromosome abnormalities: 4 with 45,X, 3 with trisomy 21, one each with trisomy 13, dup 6q. One had normal chromosomes. Two cases, in which chromosome analysis was unsuccessful, were morphologically suspected to be trisomy 13. Nine of the 12 fetuses had either bilateral cystic hygroma of the neck (7 cases) or nuchal bleb (2 cases: trisomy 13 and dup 6q). Two of the 3 remaining cases (trisomy 21) had loose skin of the neck, and one had edematous swelling of the skin of the neck. Except for the last case of trisomy 21, 11 fetuses (91.7%) had severe and/or rare cardiovascular malformations. They were divided into 3 major groups: a) spectrum of hypoplastic left heart syndrome (45,X and dup 6q), b) double outlet right ventricle, agenesis of semilunar valve (trisomy 13), and c) abnormality of atrioventricular orifice or valves (trisomy 21). One fetus with normal chromosomes had persistent left superior vena cava instead of absent right one and calcification of myocardium. Histological observation of edematous skin demonstrated the abnormal distribution of lymph vessels, including their absence. Some cases showed hypoplastic thymus. To integrate the findings of the present study and the descriptions in the literature, a pathogenesis is hypothesized in relation to migration of neural crest cells and extracellular matrix.

Quantitative description of subjective symptoms in pregnant women and their changes during the perinatal period

To study chronological changes in the spatial representation of subjective symptoms in pregnant women, the authors formulated a check list to employ throughout the period of pregnancy. Using this check list, 162 pregnant women were examined (1,773 examinations). Women who were delivered of normal full-term neonates were adopted for the study. The authors obtained the following results: The simple analysis supported the previous descriptions in many reports in literature. From multivariate analysis (biplot analysis and latent structure analysis) using no external standards, the distribution structure of the spatially represented subjective symptoms during pregnancy varied among the five periods of pregnancy (weeks 0-12, weeks 13-20, weeks 21-28, weeks 29-34 and weeks 35-). These results suggest that, for mass screening of high-risk pregnancy, a filter of subjective symptoms for each of the above-mentioned 5 periods of pregnancy should be prepared.

[Exercise capacity in ischemic heart disease evaluated by continuous wave Doppler echocardiography]

To evaluate exercise capacity in ischemic heart disease, we measured the peak aortic flow velocity by continuous wave Doppler (CWD) echo from the suprasternal notch at rest and during multiple load ergometer testing in the supine position. The study subjects consisted of 14 normal controls, 10 patients with effort angina and 21 patients with old myocardial infarction (OMI). In all the patients and in five normal subjects, thallium-201 emission CT (SPECT) was also performed immediately after exercise and four hours subsequently. Change in the peak aortic flow velocity during exercise was compared with the extent of perfusion defects and the presence or absence of redistribution images on SPECT. At rest, the peak aortic flow velocity was not different from that of the other three groups. During exercise, it increased progressively (crescendo type) as the exercise was increased in normal subjects. In patients with angina pectoris it increased at the initial stage of exercise, but was unchanged (crescendo-plateau type) or decreased (crescendo-decrescendo type) at the maximum exercise stage. Redistribution images on SPECT were seen in nine of 10 patients with angina pectoris. In patients with OMI, change in the peak aortic flow velocity was of the crescendo type in 14, crescendo-decrescendo in three and crescendo-plateau in one. In the remaining three it decreased from the beginning of exercise (decrescendo type). Redistribution images were seen in all patients with the crescendo-decrescendo, crescendo-plateau or decrescendo types. Furthermore, myocardial damage was much more extensive in the decrescendo type. We conclude that exercise continuous wave Doppler echocardiography is a useful means of evaluating exercise capacity in ischemic heart disease.