Search

An image-classification algorithm based on an alphabet of local binary patterns is described. The efficiency of the proposed algorithm is demonstrated using a test image set of handwritten digits (MNIST). A comparison of the algorithm’s properties with similar results to those of convolutional artificial neural networks is presented.

The eye resolving power and human visual acuity are the most important parameters of physiological optics. In this study, we compare our results of psychophysics, subjective measurement of visual acuity, with the results of electrophysiological measurement, objective measurement of visual acuity using vanishing optotypes that differ in spatial (sharp or non-sharp) and semantic (animate or inanimate) properties. The subject is instructed to provide consciously the classification of images: are they sharp, or are they non-sharp. At the same time the subject unconsciously provides another classification: what objects presented to him as images are animate or inanimate. This unconscious classification was provided by the subject without instruction to do it, and it was independent of his conscious decisions made during testing of the presented images of the objects (sharp or non-sharp). We see the observer’s unconscious classification (animate or inanimate) only using electrophysiological measurements. The markers of unconscious recognition are the visual evoked potential components P200 and N170, recorded from the frontal brain area. The discovered effect works as an instrument for objective visual acuity measurements.

This study is devoted to the search for electrophysiological indicators of the human perception of a threat in a complex background target environment. It is shown that the human visual system can distinguish between threatening and nonthreatening stimuli at an unconscious level, and the primary stages of image processing are similar to those during conscious perception. An increase in the amplitude of the positive wave of the evoked potential with a latency period of 320 ms and a decrease in the amplitude of the electroencephalogram rhythms at a frequency of approximately 12 Hz in the range of 350–750 ms after presentation of a threatening stimulus can serve as electrophysiological indicators of unconscious perception. The results obtained are considered from the perspective of the matched filtration model.

This paper proposes a principle for synthesizing a complex target environment and identifies the optical masking characteristics for unconscious perception of a signal. A signal hidden by masking is supplied on the periphery of the field of view for a short time interval, unconsciously activating wide-angle human “periscopic vision” that possesses low spatial and high temporal resolution. In these studies, the selective attention of the narrow-angle central-vision channel with high spatial resolution was charged with a pseudotarget. We assumed that peripheral vision is capable at that instant of unconsciously perceiving signals hidden by a mask and storing them in memory. It was established that the unconscious low-frequency descriptions of the signals stored in memory influence decision making and control the operator’s involuntary motions under conditions of indeterminacy. Opponent-style implementation of the interaction of central and peripheral vision can serve as a pattern for further refining artificial control systems.

A software–hardware complex for studying the neurophysiological mechanism that controls saccades is described and makes it possible to generate images using the motion of a window synchronized with eye movement. This complex makes it possible to demonstrate to the subject either the part of the image perceived only by the foveal region or the part perceived by the periphery of the retina.

This study investigates the markers of operation of the classification mechanism that is invariant to scale transformations of images of test objects. Observers were asked to classify images according to the criterion of animate/inanimate object. Two series of studies were conducted using the method of cognitive evoked potentials. The angular sizes of the image of the objects presented to the observer were 3° in one series and 0.4° in another. It was established that the classification of the images of the stimuli of various angular sizes according to semantic features (animate/inanimate) causes a difference in the amplitudes of the P200 components (in the leads F7 and F8 of the frontal area). The role of the P200 components of the evoked potentials in the frontal areas of the brain as a marker of the classification process was investigated. It is observed that neural networks of the prefrontal cortex use the invariant description of images implemented in the previous stages of their processing to perform the classification of objects.

The results of a study on the corrective effect of training using a visual-motor focused activity in a virtual environment on the functioning of the visual system of patients with depression are presented. The effects were evaluated using the method of cognitive visual evoked potentials. Digitally filtered images were used as visual stimuli in measuring such potentials. It was shown that in patients suffering from depression, the early stages of processing visual information are disrupted, mainly by the parvo channels, and stimulation using virtual reality systems makes it possible to correct these disruptions.

This paper discusses the characteristics of eye movement in tasks of recognizing texts and comics with descriptions of dynamic genre scenes. The interconnection of the semantic space of the comics mentally constructed by the observer and the semantic space of test images constructed by the researcher is analyzed by measuring the characteristics of eye movements, which play the role of a distinctive marker that projects onto the test object a subjective algorithm for analyzing the content. The common features inherent to the distribution of saccades and fixations of the gaze in image space when comics and textual images are being experienced probably reflect the algorithm for detecting the semantic, conceptual structure of images common to the analysis of different methods of transferring information in multimedia. This algorithm breaks down under various diseases of the brain. It is assumed that, regardless of the difference of representing information in texts and comics, understanding is provided by common mechanisms for constructing the internal, imaginary, time-varying content.

The interaction between the mechanisms of local and global image analysis at the level of the magnocellular and parvocellular channels of the human visual system was studied. Using wavelet filtering, the spatial frequency composition of texts presented to observers was varied. It was shown that gradual blurring of texts via wavelet filtering interferes with the work of the parvocellular system but simultaneously increases the contribution of the magnocellular system during reading. With an increase in the wavelet element scale, the parvocellular system receives insufficient information for effective work, and in this situation the magnocellular system determines the strategy of eye movements. In addition, the necessary frequency range that ensures the functioning of the reading process is provided.

This paper describes an algorithm for distinguishing between images of two classes of objects: artificial and natural. An approximation to the image is generated using graphical elements similar to the receptor fields of neurons in the primary visual cortex (Zone VI). We show that the approximation-error distribution for natural-object images lies at higher values of the approximation error than that for artificial-object images. This difference makes it possible to detect artificial objects against natural backgrounds.

This paper presents the results of a comparative computer analysis of monochromatic drawings by mentally healthy subjects and patients with schizophrenia. The contours of the images were distinguished in each of the drawings, after which the following characteristics were calculated: the total length of all the contours of the image, the mean size of the image, and the ratio of the total length of the contours to the mean size. The spatial-frequency spectra of the images were also computed. It is established that the mean size of the drawings and the relative length of the contours in the drawings by the patients with schizophrenia are less than in those of the healthy subjects—i.e., the drawings of the patients with schizophrenia are less detailed. Moreover, these drawings are characterized by a certain rise of the spectrum in the region of medium spatial frequencies. As shown by a computer experiment, on the basis of an analysis of twenty arbitrarily chosen drawings by the patients with schizophrenia, a positive diagnosis can be delivered in 92% of all the cases. When the same sample of drawings by healthy subjects is analyzed, the probability of an incorrect diagnosis was 6.5%. The resulting data are regarded as evidence of dysfunction when there is schizophrenia of the magnocellular system, which provides a global analysis of the images and an increase of the internal noise level of the visual system.

This article gives an analysis of the interconnection of the optical properties of incomplete (fragmented) images and the psychophysical recognition thresholds of objects as necessary conditions for the appearance of visual insight. A well-known method of measuring the perception thresholds of fragmented images—the Gollin test—is proposed for the first time for solving the problem of evaluating the characteristics of insight in a person. The architecture of the neural networks that ensure the appearance of insight is considered.

The work of an operator in solving two classification problems when working with one image alphabet is studied. From ninety visual stimuli, half of the images contained animate objects, and the other half contained inanimate objects. The first task was to classify the images according to a semantic attribute—whether they contained an animate or inanimate object. This alphabet of stimuli was then subjected to wavelet filtering in a low- and high-spatial-frequency region, regardless of semantic significance. The second task was to classify the stimuli according to a physical attribute—a blurred or unblurred object in the image. Electrophysiological monitoring of the operator’s work—recording of the induced visual potentials from the entire surface of the head—made it possible to detect that, from the beginning of the stimulation until the organization of the motor response, parallel processing of the observed signal occurs according to the different semantic and physical attributes. The responses of the temporal and frontal sections of the brain associated with the semantics of the images are distinguished, even under those conditions in which the subject’s task was to classify the physical properties of an image of an object.

It is well known that an important function of binocular vision is to increase the SNR. When a signal simultaneously arrives along two channels, it is summed if the channels are statistically independent. The binocular visual acuity can then theoretically be increased by comparison with the monocular value by a factor of 2, or 1.4. This paper gives the results of experiments to measure the visual acuity by means of Landolt rings. It is established that binocular visual acuity is greater than monocular by a factor of 1.3 on the average. It is assumed that this can be associated with internal multiplicative discretization noise at the level of the retinal receptors. This result is important for understanding image processing in different channels of the visual system.

The methods of neuroiconics and functional magnetic-resonance tomography are used to investigate the factors that limit the possibilities of visual search. The influence of an image of a human face hidden in the background on the activity of the observer’s brain was recorded during the task of tracking a moving ring. It is established that images are unconsciously perceived under threshold-presentation conditions, and this is reflected in the activation of the fusiform gyrus—a region of the brain that participates in face recognition. Under above-threshold presentation conditions, the parietal and frontal regions of the brain were also activated, but activity in this case decreased in the auditory, motor, and certain other regions of the brain not occupied in signal processing. The resulting data reveal the significance of the background semantics under conditions of visual search and explain how the unconsciously perceived optical characteristics of a background image can affect the operator’s functional state.

This paper discusses the process of perceiving dynamic images subjected to processing with spatial-frequency filters that simulate the characteristics of the receptive fields of the neurons of the primary visual cortex. A technique was used that makes it possible to give a quantitative estimate of how subjects perceive the emotional state of people’s faces on images presented to the subjects. It was shown that, besides the vertical and horizontal components of the spatial-frequency spectrum, a substantial role is played by the diagonal components in the process of perceiving the images of faces. Even though the visual system is less sensitive to the diagonal components than to the vertical and horizontal ones, the information contained in them makes it possible to distinguish the individual features and emotional state of a person’s face.

The development of methods of digitally synthesizing and processing images has made it possible to use the methods of iconics to deliberately create test images that selectively activate various structures of the visual system. The methods of processing neurophysiological data, including not only images of the activity of the entire brain but also so-called neuroimaging methods, have made it possible to discriminate the brain structures activated as a result of this selective action. The goal of this study is the spatiotemporal localization (mapping) of the regions of the brain that participate in making decisions concerning the shape of textures. It is established that a subject’s reaction time correlates with the degree of ordering of the textures and with the latency of the late components of the induced potentials in the frontal cortex. The time for a person to make decisions in the task of recognizing a specified class of textures is thereby determined. Mapping of the brain by the method of functional magnetic-resonance tomography showed that the activity of the brain in the process of making decisions involving recognition occurs in the frontal cortex of the human brain.

Psychophysical studies of the visual perception of incomplete contour images have been carried out. The images were synthesized by means of wavelets. The wavelets consisted of the difference-of-Gaussians (DoG) function and oriented wavelets whose spectrum is limited both in the frequency range and in the orientation range. The image size and the number and angular size of the wavelets were varied. The dependence between the characteristics of the wavelets and the recognition thresholds of the incomplete images was established.

Various images of the Muller–Lyer figures have been obtained by digital filtering. The filtered images contained a definite spatial-frequency spectrum, predominantly low, medium, or high frequencies. The filtering was carried out by convolution of the images with wavelets that are the difference of two Gaussoids with half-width differing by a factor of 2. The equalization threshold of the Muller–Lyer figures was measured by presenting images subjected to digital processing and without processing, thereby measuring the threshold for bringing about the illusion. The Muller–Lyer illusion was caused by all the stimuli, but it was reliably larger in response to the presentation of the image with a predominantly low-frequency component. The modelling of the Muller–Lyer illusion must take into account the spatial-frequency spectrum of the test image and the characteristics of the pass-band filtering in the spatial-frequency channels of not only the primary but also the higher divisions of the visual system, which construct the envelope based on the primary filtering of the image.