WO2004079637A1

WO2004079637A1 - Method for the recognition of patterns in images affected by optical degradations and application thereof in the prediction of visual acuity from a patient's ocular aberrometry data

Info

Publication number: WO2004079637A1
Application number: PCT/ES2004/070012
Authority: WO
Inventors: Rafael Navarro Belsue; Oscar Nestares Garcia
Original assignee: Consejo Superior De Investigaciones Científicas
Priority date: 2003-03-07
Filing date: 2004-03-08
Publication date: 2004-09-16

Abstract

The invention relates to a pattern-recognition method involving the use of a decomposition, based on channels which are tuned at different frequencies and orientations, of the original patterns and of the image observed. According to the invention, simplifications can be made to the degradation model using the aforementioned decomposition, such that, from the observed image, it is possible to calculate the probability of each of the original images generating said observation. The inventive method is Bayesian and, as such, can include statistical information about the most probable degradations and the most abundant patterns as well as about the cost of making an error in a determined pattern. In this way, the invention provides a more reliable pattern recognition method which is adapted to the requirements of each application and which can, above all, withstand possible optical degradations to the image. One application which is of particular interest to the optical and ophthalmology sector consists in predicting the visual acuity of a patient from data relating to optical aberrations of the eye, provided by an aberrometer.

Description

TITLE

PROCEDURE FOR THE RECOGNITION OF PATTERNS ON IMAGES AFFECTED BY OPTICAL DEGRADATIONS AND ITS APPLICATION TO THE PREDICTION OF VISUAL AGUDEZA FROM THE PATIENT'S EYE ABERROMETRY DATA

SECTOR OF THE TECHNIQUE

The invention is directed to all areas where automatic recognition of patterns in images is necessary, in general in automatic inspection applications using optical means, and in particular in surveillance, process monitoring, quality control, simulation applications. visual process for clinical purposes, etc. Its application is especially indicated when the observation conditions do not guarantee a good image quality. It is a numerical procedure of pattern recognition based on visual perception that can be performed using computer methods. The application to the prediction of visual acuity is directed to the area of health, specifically ophthalmology, optometry and ophthalmic optics. In this case the numerical procedure combines optical and psychophysical models of visual perception, with the procedure of pattern recognition in images, described above.

STATE OF THE TECHNIQUE

Pattern recognition in images is an area of great interest within automatic image analysis, and with multiple applications. Among them, it is worth noting the optical character recognition, the recognition of targets in military applications, classification of biological species observed by optical means, active surveillance with automatic recognition of objects of interest, etc. Of special interest in the field of ophthalmology and optometry is the prediction of the patient's visual acuity from ocular aberrometry data.

Due to the great interest of this technique, many improvements have been developed regarding the optical recognition methods of original patterns, based on optimal correlation filters. Much of these improvements have been to preprocess the observed image to correct certain factors that cause it to move away from the patterns originals, such as geometric distortion, scale changes, etc. Another set of important improvements has been aimed at trying to recognize patterns in images affected by a large amount of noise, for which it is very useful to consider probabilistic and statistical models. However, there are very few methods proposed to recognize patterns in images affected by important optical degradations and unknown a priori, and practically non-existent methods proposed to treat optically degraded images with added noise. Most optical recognition methods assume that information on optical degradation is known, and therefore can be compensated in a process prior to recognition, although there are frequent cases in which this information is not available. Therefore, it is necessary to develop specific recognition methods that are robust to unknown optical degradations in addition to noise. For example, apart from the optical degradations introduced by the image capture system, the atmosphere can introduce optical degradations that in principle are variable, random and unknown.

In this sense, previous work showed that, using methods of representing images inspired by human vision, it was possible to recognize patterns in the presence of certain optical degradations, mainly blurring [A. Vargas, J. Campos, R. Navarro (2000). "Invariant pattem recognition against defocus based on subband decomposition of the filter", Optics Communications 185: 33-40]. The present procedure is also based on methods of representation of images based on human vision, but by combining them with a more general Bayesian model, it allows treating cases in which degradations are not limited to blurring, and is also included in a way Natural noise in the model. Therefore, the present process is broader and more generic, giving rise to the present invention.

On the other hand, the measurement of visual acuity, AV, is the most widespread form of evaluation of visual quality in patients, and has long been the best and most efficient method of performing subjective refraction. That is why it has become the most practiced visual test, being carried out in a conventional manner in ophthalmological clinics, optometric establishments, and even in medicine General and legal. The AV test allows to quickly and accurately detect both retinal or neural problems (amblyopia), as well as optical defects (myopia, astigmatism, etc.). In the first case, visual acuity is only the first step for diagnosis, so that other tests such as fundus exams, etc. have to be performed. However, from the optical point of view, AV alone allows to obtain the optimal graduation of the patient, as it is very sensitive to any optical defect. Despite its widespread use for many years, there has not been a model so far to predict the values of visual acuity in specific patients, for several reasons. First, obtaining visual acuity is done by assigning a visual task to the patient consisting of recognizing the objects shown. Since the human eye, including the emmetropic, has a significant amount of optical aberrations [Liang, J. and DR Williams (1997). "Aberrations and retinal image quality of the normal human eye." J. Opt. Soc. Am. A 14 (11): 2873-2883] Conventional pattern recognition methods fail as models of the visual system since they predict a much lower AV than normal subjects, which seem to tolerate, without major problems, quantities Moderate aberrations. Therefore, the only way to predict AV is by making realistic models of the different stages of the visual process. In recent years, there have been significant advances in both optical modeling, simulating retinal images of optotype charts [A. Guirao, J. Porter, DR Williams, IG Cox (2002), "Calculated impact of higher-order monochromatic aberrations on retinal image quality in a population of human eyes", J. Opt. Soc. Am. A 19 (1): 1-9], as in the models of visual representation in the visual cortex [Landy, MS and JA Movshon (1991). Computational models of visual processing. Cambridge, MA., MIT Press]. That is why the procedure that allows the recognition of patterns in the presence of optical degradations, object of this invention, allows for the first time to face the problem of modeling and predicting visual acuity.

DESCRIPTION OF THE INVENTION The present invention consists of a method for recognizing patterns in images subjected to optical degradation and noise, from a finite and predetermined set. This set of patterns is stored in digital format, and in a gray scale (intensities between black and white) or colors. Depending on the specific application, the observed image can be acquired by means of an optical image capture system (for example, in surveillance applications), or it can be a simulation of an optical capture system (for example, a simulation of the retinal image of an object from the optical data of an eye model). In any case, the procedure is indicated for observed images that have undergone an a priori unknown optical degradation, introduced either by the capture system (camera, eye, etc.) or by factors external to it (atmospheric turbulence, etc.).

Figure 1 shows a block diagram presenting the data and processes carried out by this procedure, which are described in more detail below.

It is based on an optically degraded digital image, which may come from a scene observed by an optical image capture system and converted by an appropriate procedure to a digital image, or be the result of a numerical simulation of said capture processes . This digital image will be compared with the digital images of the default pattern set, using digital computers.

The degraded image is transformed by applying a multiscale / multiorientation filter bank to obtain a visual representation of it. This same procedure is applied to the images containing the set of preset patterns, being able to execute this process in a previous time and recovering directly the visual representation stored in a suitable device. The method is flexible in terms of the type of filter to be used (Gabor, Gaussian derivatives, Laplacians, etc.), number of filters and arrangement of the scales and orientations, which allows it to be adapted to the specific needs of each application.

Next, the probability of having generated the observed image is calculated for each pattern of the preset set. For this, a Bayesian method is applied that makes use of the visual representation of the images, and in which an approximation of the unknown degradation that is affecting the observed image is implicitly estimated. Using this visual representation we introduce the further simplification that the frequency response of the unknown optical degradation is constant within the frequency range that each channel lets through. This simplification makes us move from an undetermined system to a particular system, and it is possible to calculate the probabilities of having generated the observed image. With the previous assumption it is possible to formulate the following simplified observation model for the versions of the observed image filtered with each of the Gabor filters of the representation scheme:

or, (x) = (/ 2 (x) * c (x)) * g, (x) + η, (x) _a¡ A, c _; (x -u,) + η, (x), z = l, ..., N _c (1)

where o, (x) is the observed degraded image filtered with the z ^" -th Gabor filter, g ₍ (x), and contaminated with the additive noise η, (x); h (x) is the impulse response of the unknown optical degradation; and c (x) is the image that contains the original pattern without degrading.Using the above simplification, you get to the right part of the previous equation, where c, (x) is the image input pattern without degrading and filtered with Gabor filter z ^'-th; _(h:, or are constant multiplicative factor and the overall displacement which is approximately the frequency response of the optical degradation in frequency range allowed by the z ^' -th channel; and, finally, N _c is the number of Gabor channels of the representation.

With the previous observation model we can formulate the posterior probability of the original character c, together with the parameters that model the optical degradation

{/ z ,, u,}, given the set of observations {o,}. Applying the Bayes rule we obtain the following expression, which makes explicit the possibility of including a priori information, such as the probability of a specific pattern, or as the a priori probability of the parameters that model optical degradation:

_J p (c, {A _I , u _I } | {o _l }) = .S: ^ ({ _l } | c, {A _l , u,}) _j p (c) / 7 ({Λ _l , u _l }) (2)

where for convenience of notation we have expressed the images as vectors; K is a noπnalization constant. The a posteriori probability is equal to the likelihood, or conditional probability of the observations given the model parameters, multiplied by the a priori probability of the model parameters. We have supposed that the original input image and the parameters that model the optical degradation are independent. If there is no prior information, it is possible to assume that the parameters that model the optical degradation are equiprobable, resulting:

| {o,}) = K'p ({o _i } | c, {2 _{! 3} u _; }) (c) ₍₃₎

where K 'is another normalization constant. The maximum a posteriori estimator, MAP, for the input pattern, c, and for the parameters of the optical degradation, {, ü,} is the one that maximizes the posterior probability of equation (3):

(c, {/ 7,, ü,}) = arg max p ({o,} | c, {/ ?,, u,}) _/ ? ((:) ₍₄₎

(c, {Λ „u,})

where the likelihood function ¿> ({o,} | c, {z ,, u,}) is given by the noise probability density function, according to the observation model of equation (1). Assuming conditional independence between channels, as well as between pixels, the likelihood function results:

N _j p _r _({o;} | c, _{2;, _u;}) = fÍfJ _τl _(σ; (x) - / _2, c, (x -u)) (5) ι = l

Next, we incorporate the a priori probability of the original undegraded pattern, c, which is determined by the fact that the input image must be co-sponged with some of the patterns in the pre-set pattern set. In this way, this a priori probability can be expressed as a sum of delta functions each associated with a pattern, with a weight given by the a priori probability of that pattern appearing in the image. Considering all the equiprobable patterns, the posterior probability is:

where {c ⁷ } ^, are the images corresponding to the N patterns. The introduction of this a priori probability means that the space of all the possible configurations of the intensities in the input image, resulting that the probability is nonzero only for ce {c ^J } *,, points at which the posterior probability is:

/ ⁷ (c = c ^J , {Λ _l .u _I } | {or _I }) oc πf | / ⁷ _{T] (} (or _I (x) - A _I c, - '(x - «,)) ( 7)

; = 1 x

Thus, Bayesian recognition consists first of all in choosing the degradation parameters that maximize the probability in (7) for each pattern in the set, and then choosing as the recognized pattern the one with the highest probability, which is precisely the one corresponding to the global maximum of the probability a posteriori. Obtaining the parameters {Λ ,, ü _; } maximizing the expression (7) can be done individually for each channel, and then multiply the maximum values for each channel to obtain the probability. For a particular channel, i, and assuming Gaussian white noise, maximizing the probability is equivalent to minimizing the following error function:

/ = ∑ (*, (x) - / (x-,)) ² (8)

It is possible to demonstrate that the function (8) is minimized for the value ü / that maximizes the correlation function Cθ7τ (u ₍ ) = ∑σ, (x) c (xu,), and then

fy = Corr _ι ^J (Ü ^J _¡ ) / K? , with K, ^J = ∑ ( ^c / ( ^x ) j. Thus, the value of the probability a

later for the employer and it turns out:

P _j = max (9)

The result of the Bayesian method is a probability like the previous one, associated with each of the patterns and the set. This information can be used in many ways, depending on the application. One of the most interesting possibilities is to select the pattern with the highest probability as the recognized pattern from The observed image. It is also possible to reject the hypothesis that some of the patterns are present in the image, if confidence thresholds are not exceeded in the calculated probabilities. Other additional information provided by this method is an estimate of the most likely degradation parameters, which can be used to recover an approximation of the unknown optical degradation that affected the observed image.

In summary, the result of the application of this procedure provides:

1.- A set of probabilities that the pattern that appears in the degraded image corresponds to each stored pattern, being possible to establish an ordered list of patterns from highest to lowest probability.

2.- The pattern among the set of patterns that most likely generated the observation. This is the answer provided by the Bayesian model from maximum to posterior. 3.- An estimate of the most likely optical degradation that affects the observed image.

The present invention could be applied in a variety of practical situations, including: 1. Optical character recognition (known by its acronym in English, OCR), in blurred or degraded images.

2. Observation of objects in the sky from fixed terrestrial platforms, using optical means, in such a way that to the possible degradations introduced by the optical instruments (blurring, etc.) those added by atmospheric turbulence are added. Specific examples of application are the recognition of birds, aircraft (for example, airplanes and helicopters), satellites, stars, stellar objects, etc.

3. Observation of objects, both in the sky and on the ground, using optical means from mobile, terrestrial or aerial platforms, that is, images of mobile objects captured from also mobile platforms, so that degradations due to movement occur (in addition from those of optics and the atmosphere). Application examples are the recognition of numbers and letters on vehicle license plates by images taken from surveillance helicopters, recognition of military targets in images taken from vehicles or reconnaissance aircraft, etc.

4. Images of biological specimens, captured by microscopy or other biomedical imaging techniques, and affected by degradations introduced by the turbidity of the biological environment, preparations, etc., as well as by the formation and image capture systems, in which It is necessary to recognize the specimen to proceed with its classification.

5. Prediction of the visual acuity of a patient from the data of the optical aberrations of the eye provided by an aberrometer. This is the exemplary embodiment proposed below (see Figure 2).

EXAMPLE OF EMBODIMENT OF THE INVENTION

As an example of implementation, it is proposed to implement a procedure for predicting the visual acuity of a patient in an aberrometer. Current aberrometers usually have data analysis software, some of which even simulate the appearance of a letter or optotype on the patient's retina. In this case, it is about adding an additional software module, so that the aberrometer can provide the prediction of the patient's visual acuity. This prediction is useful, since if it is checked against the visual acuity obtained in the clinic, if there is a significant disagreement, this can be interpreted as a symptom that the patient may have a deficiency in their visual perception.

The specific procedure is shown in Figure 2. It is based on input data, which is used to establish a personalized model of the patient's eye. This model consists of an optical part, a retinal part consisting of cones sampling, and a neuronal representation of the image, which are applied sequentially. The optical model starts from the optical aberrations to obtain the optical transfer function (OTF) that acts as a linear filter on the input test image. The OTF filter is modified to incorporate the effect of sampling retinal photoreceptors (cones in photopic vision and rods in scotopic vision). The second part of the model is applied to the filtered image consisting of a Pyramidal multi-scale / multi-orientation decomposition through a filter bank (Gabor, Gaussian derivatives, steerable pyramid, etc.) In this case, a Gabor filter bank has been chosen [O. Nestares, R. Navarro, J. Portilla, A. Tabernero (1998), "Efficient spatial-domain implementation of a multiscale image representation based on Gabor functions", J. Electronic Imaging, 7; 166-173], which is followed by normalization by the low-pass residue to pass to contrast units. Filter frequencies are set so that the maximum frequency matches the standard models. This decomposition constitutes a schematic but realistic model of the representation of the image in the visual cortex. Finally, a contrast threshold is applied, so that contrast values that do not exceed the threshold are not considered. In this way the contrast sensitivity of the visual system is modeled. The complete model can be applied to any type of image, giving rise to a cortical representation of the same, and which in turn constitutes the entrance for the pattern recognition procedure that must be robust, or present some invariance, against to the presence of optical degradations (aberrations). The output is the character (optotype) of the alphabet that most likely corresponds to the input image. The entire procedure is applied to a set of images of input optotypes simulating the clinical procedure to obtain visual acuity.

Optical model

The optical model is determined by the wave aberration, which in this case would be described by the coefficients of a development in Zernike polynomials provided directly by the aberrometer, and by the parameters that describe the Stiles-Crawford effect of the patient. This effect is equivalent to an apodizing filter described with a Gaussian, of a certain width σ and centered on certain coordinates in the plane of the pupil. In this example, it has been assumed that we do not have the specific data of the patient's eye available and standard average data is used (in the case of the Stiles-Crawford effect, it is not critical to have accurate patient data, so using average data extracted of literature is usually a good enough approximation). From these data describing the wavefront in the pupil, the OTF optical transfer function is obtained, which is the autocorrelation of the wavefront in the pupil. The retinal optical image of an input test image is obtained by a filtering operation in the spatial frequency domain, the OTF being the filter. In this example, a monochromatic case has been considered, although its extension to the polychromatic case is immediate, if the optical aberrations for various wavelengths in red, green and blue are known. In this case, our method consists in applying the corresponding OTFs to each of the chromatic channels of the RVA input image for those 3 wavelengths, and obtaining the retinal images for the 3 chromatic components, then a transformation is carried out to pass to the representation in CIELAB chromatic coordinates that best model the color behavior of the visual system. From here, the image corresponding to brightness L is used for the rest of the procedure. The OTF is modified due to the spectral overlap produced by the sampling of the photoreceptors. In this example the case of photopic vision has been considered, so the sampling is given by the distribution of cones.

Input data for models

The input data are of great importance for the realization of the model, since they are the ones that will characterize that particular patient or eye. We can divide them into optical and cortical data, as they correspond to one part or another of the complete model. However, in a more practical way, we divide them according to their availability in the specific patient. That is, when we have the specific data measured in the patient (1), or otherwise, we use a model with standard data (2). In Figure 3 it has been assumed that the only data available are those of aberrometry, in which case the prediction will be more reliable in patients whose retina and visual cortex are normal and therefore do not present any conditioning in this regard.

Test images and optotypes

To obtain visual acuity, test images containing calibrated optotypes are introduced so that their sizes correspond to specific values of visual acuity. The image is analyzed by applying the pattern recognition to each optotype, assigning a value of 1 or 0 (or Boolean variables true or false) in case of success or failure, respectively, in the recognition. A threshold is established, for the number of failures allowed, to override the task for a certain optotype size, corresponding to a certain visual acuity value. Both the optotypes and the threshold of hits will be similar to those used in the procedure to measure the visual acuity used in the concrete reality or clinic (a typical percentage of hits is at least 75%). Figure 2 shows an example of a test image consisting of 4 rows, each containing as many characters (optotypes). The size of the characters in each row corresponds to a certain value of visual acuity. Thus, the full size of the character is 5 times the thickness of the stroke, and this in turn corresponds to the value of visual acuity. In this case a decimal scale has been used, such that visual acuity unit corresponds to a stroke size that subtends a minute of visual field according to the optical model. In this example the lines of the test image correspond to visual acuities of 0.6, 0.8, 1 and 1.2 respectively (see figure).

In this example the characters are designed on purpose to meet the above specifications. Specifically, a reduced 16-character alphabet is used for several reasons. On the one hand, visual acuity tests do not use all the characters of the alphabet, having a predilection for a subset, and on the other hand, reducing the number of possible characters to 16 saves the calculation time in the recognition stage. In any case, the alphabet can be chosen so that it is identical to that used in the actual procedure used in the specific clinic, as already mentioned.

Pattern recognition and visual acuity

Pattern recognition is done by extracting the portion of the image resulting from applying the models contained in each of the optotypes. The procedure consists of several stages:

- The procedure of recognition of patterns affected by optical degradations is applied, comparing the image of the optotype, with the 16 optotypes of the alphabet used. This procedure returns as an answer the optotype among the 16, most likely to be the same as the input.

- When the optotype provided by the method matches the optotype of the test image, it is considered a success and failure otherwise. In this alphabet, the chance of success by simple chance is 1/16. - A value of visual acuity is considered exceeded, when the number of hits on the corresponding line is equal to or greater than 75% (ie 3/4 or

4/4).

The procedure begins with the upper line (major scale, standard visual acuity, that is 1). If the line is overcome (at least 75% of hits) it is passed to the next higher line (1.2). In case of having more failures of the peπnitidos, it goes to the bottom line (0.8). The procedure is stopped if in ascending trajectory the threshold of successes is not exceeded, or in descending trajectory when it is exceeded, returning as the value of visual acuity that of the last exceeded line.

The procedure can be monocular or binocular. In the second case, the procedure consists in properly combining the results of both eyes. This can be done either by a final stage in which the results of both eyes are checked to eliminate errors, or by merging the visual information contained in each of the images, optimizing the result.

The result of the application provides:

1.- The simulated optical image of any input test image according to the personalized optical model of the patient's eye.

2.- The cortical representation with the format of a multiscale / multiorineration pyramidal decomposition, which includes the effect of the optical degradations of the test image.

3.- Success or failure in the task of recognizing each of the objects contained in the input image. The procedure operates on a set of possible objects

(opotypes) predefined.

4.- Visual acuity predicted in the patient.

Claims

1. Numerical procedure to recognize patterns in images subjected to generic and unknown optical degradations, from a set of preset patterns. Optical degradations may come from the physical process of image acquisition, or from numerical simulations of said physical process.

2. A method according to claim 1, which uses a system for representing images, both observed and patterns, multiscale / multiorientation implemented with a band-pass filter bank. This system is flexible both in the filter field (Gabor functions, Gaussian derivatives, etc.) and in the arrangement of the center frequencies and their bandwidth, and these parameters can be adapted to the specific application.

3. Procedure, according to claims 1 and 2, which calculates the probabilities that each pattern of the alphabet has generated the observed image, using a method

Bayesian, and using the representation system described in claim 2. This method has the particularity of allowing a priori information to be included on the statistics of the appearance of the patterns, as well as the statistics of the parameters that model the optical degradation.

4. Procedure for calculating Bayesian probabilities, according to claim 3, characterized by (a) based on the value of the correlation between the observed image and the pattern; (b) because the correlation is applied in the different subbands of scales and orientations; (c) transform the correlation values into probabilities and combine them according to the Bayesian method, so that the final result is a probability, with the characteristics of minimizing false alaπnas; (d) give as an answer that pattern of the set of patterns for which the maximum probability value has been obtained.

5. Pattern recognition procedure, according to claims 1 to 4, capable of recognizing patterns in images subjected to optical degradations caused, either by blurring of optical pick-up instruments, or in images captured through the atmosphere and therefore affected of the random aberrations introduced because of the turbulence. Examples of applications in which this type of degradation occurs include the optical recognition of characters in blurred images, recognition of birds in flight, aircraft, satellites, stars and stellar objects.

7. Pattern recognition procedure, according to claims 1 to 4, capable of recognizing patterns in images of mobile objects captured from also mobile platforms, so that, apart from other sources of optical degradation, they are affected by degradations due to movement during sensor integration time. Examples of applications in which such degradations occur include the recognition of license plates of moving vehicles, captured from moving platforms, and the recognition of military objectives in images captured from mobile recognition platforms.

8. Pattern recognition procedure according to claims 1 to 4, capable of recognizing patterns in images subjected to simulated optical degradations, in order to objectively assess the response of optical capture systems as a function of the level of degradation.

9. Pattern recognition procedure, according to claims 1 to 4, capable of recognizing patterns in images of biological specimens, captured by microscopy or other biomedical imaging techniques, and affected by degradations introduced by the turbidity of the biological medium, preparations, etc. , as well as the formation and image capture systems.

10. Procedure for the prediction of visual acuity in patients, characterized by the fact that they are based on realistic optical and visual image processing models based on optical abetrometry data and use the robust pattern recognition procedure for optical degradation of the image, according to claims 1, 2, 3, 4 and 8, and which generates a plausible visual model.

11. Method based on claims 1, 2, 3, 4, 8 and 10 characterized by modeling the first stages of the visual process by combining an optical model that simulates the optical image on the retina of an input test, from of the Wave division includes spectral overlap produced by retinal sampling and applies a pyramidal decomposition by means of a multiscale / multiorientation model of said image.

12. Method according to claims 1, 2, 3, 4, 8, 10 and 11, specific for the case of polychromatic lighting, characterized by (1) using as a starting data a color test image (with its RVA components) and its wave aberrations for the corresponding wavelengths; (2) simulate the retinal images corresponding to the 3 RVA colors; (3) transform to CIELAB chromatic coordinates that best model the color behavior of the visual system and (4) use the image corresponding to the chromatic coordinate L (luminosity) for the rest of the procedure.

13. Method according to claims 1, 2, 3, 4, 8, 10 and 11 characterized by: (1) the model includes a normalization through the low-pass residual channel, so that the output is given in contrast units and ( 2) a contrast threshold is applied so that values below the threshold are ignored in the rest of the procedure to simulate neuronal noise.

14. Procedure, according to claims 10 and 11, of generation and use of input test, characterized by (1) containing rows of optotypes of different sizes, calibrated according to the values of visual acuities to be checked; (2) use optotypes of test images taken exclusively from a set, or alphabet in the case of using letters; and (3) compare each optotype with the whole set or alphabet in the recognition stage.

15. Procedure for obtaining visual acuity according to claims 1, 2, 3, 4, 8, 10, 11, 12, 13, 14, characterized by: (1) considering a value of visual acuity exceeded, when the number of successes obtained in the set of test optotypes of the size coresponding to said acuity is equal to or greater than a predetermined threshold value, which is usually 75% or similar; (2) The procedure begins with the upper line (major scale, standard visual acuity, that is 1). If the line is exceeded (at least 75% of hits) it goes to the next higher line (1.1). In case you have more failures of the allowed, it goes to the bottom line (0.9). The procedure is stopped, if in ascending trajectory the threshold of successes is not exceeded, or in descending trajectory when it is exceeded, returning as the value of visual acuity that of the last exceeded line.

16. Procedure for obtaining binocular visual acuity, according to claims 10 to 15, characterized by combining the responses of both eyes to eliminate false alarms, thus improving the number of successes with respect to the monocular case, analogously to claim 5, point (3).

17. Application of the method, according to claims 10 to 16, to be used in an ocular sound meter to provide a prediction of visual acuity from the measured data of ocular sound level.