Computer implemented method for generating an image of a surface

Abstract

The present invention is related to a computer implemented method for generating an image of a surface departing from one or more fine images and one coarse image of the surface. According to a preferred embodiment, the method combines a plurality of fine images and, one or more coarse images captured at different time instants generating a new fine image at a different time instant. Captured images are preferably captured by one or more satellites.

Description

[0001]COMPUTER IMPLEMENTED METHOD FOR GENERATING AN IMAGE OF A SURFACE DESCRIPTION FIELD OF THE INVENTION The present invention is related to a computer implemented method for generating an image of a surface departing from one or more fine images and one coarse image of the surface. According to a preferred embodiment, the method combines a plurality of fine images and, one or more coarse images captured at different time instants generating a new fine image at a different time instant. Captured images are preferably captured by one or more satellites. PRIOR ART One of the technical fields with a more intensive development is the development of information acquisition techniques using satellite-based sensors, in particular images. This is the case, for example, with the use of cameras with multispectral sensors. The same is true when data is acquired from other types of vehicles such as unmanned vehicles or drones. These have the possibility of placing the capture point at high altitude, which allows image captures from privileged viewpoints since they have visual access over a very large area. In the case of satellites, the captured images are then transmitted to a ground station where they are stored and further processed. The same happens when the satellite is orbiting a planet other than the Earth or another natural satellite such as the Moon. While satellites have a privileged viewpoint, they also have several limitations in capturing images over a specific region of the target surface. One of the limitations is the specific path that the satellite has since this path cannot be modified. When the satellite is not geostationary the capture over a certain region must be done at the moment where its path is closest and this may result in a different pose for photographies taken over the same region at different time instants. Pose means the orientation and position of the focal point of the photograph captured by a camera. Each pixel of the generated image is the result of light coming from the captured surface at a position identified through the geometric line resulting from the intersection between the focal point and the image sensor pixel. If the focal point is located elsewhere, the same region is shown with the distortion due to the different pose. The time instants can be hours or typically days. To compare images with the same pose it is necessary to apply mathematical transformations on the image by applying functions that distort the shape of the image taking into account the pose of the captured image and the pose that is intended to be used for example to compare images. Another limitation of images captured by cameras installed on satellites is their extension. To obtain an image capture over a pre-established area it is necessary to combine more than one image, in particular images in which at least one region is within the region of interest, until there are enough images to cover the entire region of interest. Another no less important limitation is the presence of clouds or environmental factors that make some regions of the captured image not useful because they do not represent the captured area. In these cases, it is necessary to identify the pixels that do not provide useful information so that they are not used later. Even when the problems of posing, cropping or combining images to cover an area of interest are solved with known techniques there remains the problem of the availability of high-resolution images at instants of time when it is not possible to capture one or more images even though a low resolution image is available. The present invention solves the problem by means of a technique in which images acquired at different time instants are fused to obtain the image at the instant of interest. DESCRIPTION OF THE INVENTION Throughout this description we will consider images that have a fine resolution and others that have a coarse resolution. Fine resolution can be identified as high resolution and coarse resolution can be identified as low resolution. In the context of this invention it is not relevant how fine the high resolution is or how coarse the low resolution is, what is relevant is that there are two distinct resolutions and that the fine resolution is greater than the coarse resolution. Likewise, in the context of the invention there are two relevant operations, an aggregation operation and a disaggregation operation. The aggregation operation is to be interpreted as a transformation operation that generates a coarse image from a fine image. The aggregation operation assesses the value of each pixel of the new coarse image by combining a plurality of pixel values of the fine image. The disaggregation operation is to be interpreted as a transformation operation that generates a fine image from a coarse image. The disaggregation operation assesses the value of each pixel of the new fine image from one or more pixels of the coarse image, preferably by interpolation. The present invention, according to a first aspect, is a computer implemented method Throughout the description, only with the intention of showing more clearly those data that are received in the computer system, mainly the captured images, and those data that are calculated, the calculated images, we will make use of cap letters to identify these seconds. It is possible that an image received by the computational system is pre-processed. Since it is this image that is received by the computational system, it will be identified without a cap. Throughout the description ^^will be identified as the time instant at which the image is to be generated and the time instants ^^, ^^, … , ^^the time instants at which fine images are available. The most general case is the case where only a single fine image is available at time instant ^^. The set of time instants ^^, ^^, … , ^^in which images are available will be called the "base period" and ^^will preferably be the time instant closest to ^^. Time instants ^^, ^^, … , ^^of the base period will be denoted by sub-index ^. The first two stages of the method consist of receiving the captured images. At least one fine image at time instant ^^and one coarse image at time instant ^^, the instant at which the target fine image is generated. The method, further comprises the steps: Throughout the description we will use names associated with images such as ^^^. These images will appear either with the name alone ^^^or including in parentheses the indication of a certain pixel, i.e.: ^^^(^^). The image is the same, but the second notation ^^^(^^) refers to a specific pixel ^^of the identified image. Pixels of coarse images will be generically denoted as ^^and pixels of fine images will be generically denoted as ^^. The method processes pixel by pixel the captured reference coarse image (^^^). For each pixel a vicinity region (V) is defined which is a set of pixels formed by the pixel being processed ^^and a set of pixels around it defined by a predetermined pattern. Preferably this pattern will be described according to embodiments of the invention as a rectangular window centered on the reference pixel ^^but other patterns non- rectangular or not compact may be used as further embodiments. In the processing there are at least two coarse images, the image acquired at the instant ^^, captured reference coarse image (^^^), and an image which we will name accumulated coarse image (^^^^^). In the most general case where only a fine image is received at instant ^^the accumulated coarse image (^^^^^) is the first coarse image (^^^^) resulting from aggregating the first captured fine image (^^^). When the method receives a plurality of fine images, it will be described later below how to determine the accumulated coarse image (^^^^^). The data of both coarse images in this step are already known and there is a bijective correspondence between the pixels of one and the other image. In the corresponding pixels the same vicinity region (V) is defined so that a linear regression is established by using the pixel by pixel data pairs of the two images, the captured reference coarse image (^^^) and the accumulated coarse image (^^^^^), and determining the coefficients of the linear regression. For each pixel ^^the two coefficients of the linear regression, ^^(^^)and ^^(^^), have been determined. From these two values per pixel two images are generated, a first image, the first coarse coefficient image (^^^), having a coarse resolution and representing the coefficient ^^(^^)in each pixel ^^and, a second image, the second coarse coefficient image (^^^), having a coarse resolution and representing the coefficient ^^(^^)also in each pixel ^^. The method, further comprises the steps: Now the coefficients ^^(^^) and ^^(^^) are stored in two images and can be processed as such, in particular by transforming them into two fine images by increasing their resolution. The last stage of the method generates the target fine image (^^^^) at instant ^^by modifying the nearest fine image, the fine image ^^^at ^^if there is only one fine captured image or otherwise the accumulated fine image ^^^^^, establishing a correspondence through the linear regression now defined by means of the disaggregated coefficients. That is, the method considers that the temporal variations in the coarse image are the same as in the fine image only that the information between the coarse and fine image is transferred by disaggregation of the images storing the coefficients. The accumulated fine image ^^^^^is a fine image generated by accumulating information of the set of ^ captured fine images ^^^, ^ = 1, … , ^. The result is a fine image, (^^^^), at instant ^^at which initially only a coarse captured image (^^^) was available. According to an embodiment, the method further uses the calculated residual ^(^^)of pixel ^^wherein - - - Further calculating residual ^(^^) per pixel in the linear regression and using it in the generation of the new fine image provides a more accurate fine image. In this specific case a third coarse image is generated storing in each pixel ^^the residual, such variable being disaggregated for increasing the resolution in order to include the residual when building the fine image, now according to wherein now ^^refers to each pixel of the fine images, the generated fine image (^^^^), the images related to the two coefficients, ^^^(^^)and that is, the first coarse coefficient image (^^^) and a second fine coefficient image (^^^), and also the new third fine coefficient image (^^^). According to an embodiment that may be applied to any of the previous disclosed embodiments, the vicinity region (V) comprises a set of pixels located in a window In this embodiment, the vicinity region (V) is a rectangular region centered in pixel ^^. In a more specific case the vicinity region (V) is quadrangular centered in pixel ^^. That is, ^^= ^^. According to an embodiment that may be applied to any of the previous disclosed embodiments, the method further comprises: - - - wherein ^(^^, ^^)is a binary mask indicating which date ^^contains the optimal observation of pixel ^^and, ^^^(^^) denotes the correlation between the set of pixels of the vicinity region (V) of the reference coarse image (^^^) and the set of pixels of the vicinity region (V) of the said generated coarse image (^^^^), the correlation - - That is, the computer system now receives a plurality of ^ fine images (^^^,…, ^^^) of the same fine resolution captured at different time instants ^^^ = 1, … , ^ . The accumulated coarse image (^^^^^) has information of the plurality of fine images. Fine images are transformed into coarse images by aggregating each of the captured fine images. The accumulated coarse image (^^^^^) is calculated from the aggregated coarse images pixel by pixel wherein each pixel of the accumulated coarse image (^^^^^) is a pixel of a selected coarse image, that coarse image having the highest correlation between the set of pixels of the vicinity region (V) of the reference coarse image (^^^) and the set of pixels of the vicinity region (V) of the said generated coarse image (^^^^). ^(^^, ^^)is generated as a binary mask in the coarse level and it allows to select at each pixel ^^the pixel value of one image among the plurality of images involved under the summation sign ^(^^, ^^) ^^^^(^^). Additionally, ^(^^, ^^) is generated in the fine level from ^(^^, ^^)by disaggregation but, ^(^^, ^^)is also a binary mask. That is, if the disaggregation of ^(^^, ^^)is made by using interpolation, then each value ^(^^, ^^)is taken as 1 or 0, the closest binary value to the interpolated value. According to another embodiment, an specific interpolation uses the closest pixels of a certain vicinity in an index image. Each pixel takes the most represented value in the vicinity defined around that pixel, thus adopting either a 1 or a 0. According to an embodiment, the fourth coarse coefficient image (^^^) is a multichannel image storing the values of the tensor. According to an specific embodiment, the fourth coarse coefficient image (^^^) is efficiently stored in a single channel image wherein each pixel stores the index of the most representative image. This indexing method is equivalent to select the image that would be multiplied by 1 according to ^(^^, ^^). The same applies to the use of ^(^^, ^^)in the fine level. This condition may also be disclosed using an alternative notation. The selection of the optimal information in the accumulated coarse image (^^^^^) is carried out by finding for each aggregated coarse image (^^^^), ^ = 1, … , ^ the image of the base period ^^, ^^, … , ^^most similar to the reference coarse image (^^^) at a time instant ^^. The degree of similarity between the images is calculated by the linear correlation coefficient in the vicinity region (V) around pixel ^^as follows: where ^^^ ^^^(^^)^ is the covariance between the ^^^^(^^)and ^^^(^^)values in the window around pixel ^^, ^^^ and ^^^ ^^^^(^^)^ are the variances of the ^^^^(^^) and ^^^(^^) values in the same window respectively. That is, if for example ^^is the pixel corresponding to row ^ and column ^ of the image, a square window of ^ pixels defining the vicinity region (V) correspond to the submatrix corresponding to the indices [[{^ ∗ ^^|^^∈ ℤ ∧ ^^∈ [−^, ^]}; {^ ∗ ^^|^^∈ ℤ ∧ ^^∈ [−^, ^]}]. In this way it is possible to determine the optimal information as where ^(^^, ^^)is a binary mask indicating which date ^^contains the optimal observation of pixel ^^and ^ is the entire range of the base period ^ = {^^, ^^, … , The temporal variation between the base period and the objective date may be expressed as: and the last linear regression when generating the fine image (^^^^) may be expressed as both expressions including the residual ^ respectively when they are available according to a specific embodiment. In the last two expressions, the accumulated coarse image (^^^^^) is identified as ^^)^^^^ and the accumulated fine image ^^^^^is identified as ^^)^^^ wherein ^(^^, ^^)is determined according to a specific method. The specific method stores the binary mask ^(^^, ^^)as an image, a multichannel image since ^(^^, ^^) defines an image per time instant ^^. A multichannel image is an image formed by layers where each layer includes an additional index, in the case of ^(^^, ^^)the additional index is ^^. One way of representing a multichannel image is by means of three-index tensors. An example of a multichannel image is an RGB image where red, green and blue are in different layers. One way to identify a channel is by the term “band” since the channel represents an image with the intensities captured in a certain wavelength range ( for each color red, green and blue). The wavelength range of a band is mainly determined by the sensor used to capture the image. A multispectral sensor is an example of a sensor capable of producing multichannel or multiband images. The resulting coarse multichannel image is disaggregated generating a new fine multichannel image wherein each channel is interpreted as storing the values of ^(^^, ^^)at a determined time instant ^^. The disaggregated ^ is the new binary mask used for determining the accumulated fine image ^^^^^. According to an embodiment that may be applied to any of the previous disclosed embodiments, generating a fine image from a coarse image by disaggregating said coarse image comprises: - disaggregating each pixel of the coarse image into a plurality of fine pixels of the fine image; - the pixel values of each plurality of pixels of the disaggregated pixels of the fine are representing the captured value at a predetermined location of the pixel and, its value is calculated by interpolating pixel values of the coarse image also representing the captured value at the same predetermined location of the pixel. According to this embodiment, each single pixel of the coarse image is expanded into a plurality of pixels of the disaggregated fine image and the new values of the fine grid are interpolated from the pixel values of the coarse image taking as reference location the same as in the fine image. The reference location is predetermined. According to a specific embodiment of the previous embodiment, disaggregated pixels of the fine image generated from a pixel of the coarse image are different from disaggregated pixels of the fine image generated from a different pixel of the coarse image. In this case, each pixel of the coarse image causes the generation of a natural number of pixels in the fine image. According to a specific embodiment that may applied to any of the two former embodiments, the predetermined location of the pixel is the center of the pixel. This embodiment allows to use the location of the associated coordinates of the surface and the pixel defines a rectangular area around said location depending on the resolution of the image. According to an embodiment that may be applied to any previous embodiments, the spectral resolution of the reference coarse image is adjusted according to a method that will be identified as normalization. This normalization method also provides images that are co-registered. However, this normalization method may be used as an independent solution and claimed in an independent claim. The normalization method comprises the following steps: - capturing a coarse image (^^^) at each time instant ^^^ = 1, .. , ^ of the surface wherein the coarse image (^^^) has a predetermined resolution lower than the resolution of the first fine image (^^^). In this embodiment, the computer system receives two sets of images, a first set of fine images and a second set of images in the base period. A pair of fine and coarse images for each time instant ^^, ^ = 1,2, … , ^ of said base period{^^, ^^, … , ^^}. and further executing by a computer system the steps: - generating a representative fine image ^^wherein each pixel value of the representative fine image ^^is calculated as a representative value of the corresponding pixels values of the set of the ^ fine images (^^^,…, ^^^); - generating a representative coarse image ^^wherein each pixel value of the representative coarse image ^^is calculated as the representative value of the corresponding pixels values of the set of the ^ captured coarse images (^^^,…, ^^^); - aggregating the representative fine image ^^to an aggregated representative coarse image ^^^. The representative fine image ^^and the representative coarse image ^^are images which represent the behavior of the fine and coarse image series respectively and this can be done in various ways: with an average, with a maximum value, with a weighted sum, etc. In a preferred embodiment the pseudo-median (that coincides with the median if the data are odd, otherwise the first central value is chosen) has been chosen because in this way the chosen representative values are values that exist previously in the image series, either fine or coarse, without new values being calculated. According to an specific embodiment, the set of ^ fine images (^^, ^^, … , ^^), the set of captured coarse images (^^^, ^^^, … , ^^^) and the resulting representative fine image ^^and the representative coarse image ^^are multichannel images. In this case each channel of the representative images (^^, ^^) is the result of selecting the representative value for each of the channels of the corresponding series of images, the set of fine images or the set of coarse images. That is, at this stage each channel is processed separately from other channels. At this step, in the coarse level there are two coarse images, the representative coarse image ^^directly calculated from the coarser series of images and, the image ^^^ obtained by aggregating the representative fine image ^^into a representative coarse image. According to this embodiment, from these two coarse images the method further comprises the steps: - calculating the anomaly of ^^^as ^^^= ^^^− ^^, the anomaly of ^^^as ^^^= ^^^− ^^and, the anomaly of ^^^^as ^^^= ^^^^− ^^, being ^^a pre-specified time - - The three anomalies are three images wherein each pixel is the difference between the image used for calculating the anomaly and, the representative coarse image ^^. Now, once a pair (^^^, ^^^) of percentiles is selected, a selected set of pixels of the anomalies is also selected. Only those pixels selected by the pair (^^^, ^^^) of percentiles is used in the linear regression determining scalars ^^^^^^^, ^^^^ and ^^^^^^^, ^^^^. It is noted that this linear regression is calculated only with the anomaly of one single image, the image at ^^being preferably ^^and not for all images. Using ^^is expected to provide more accurate results. If images are multichannel images, the set of pixels selected as being those ranging between the percentiles (^^^, ^^^) are selected by using, or one single channel, a pre- defined channel, or using a new channel generated by aggregating two or more of the available channels. In this case, the set of selected pixels is kept when computing the linear regression of ^^^^^^^, ^^^^ for a predetermined ^ value in the rest of channels but, coefficients and residual are recomputed for every channel. That is, ^^^^^^^, ^^^^is computed channel by channel for the selected pixels. According to this embodiment, the method further comprises the steps: - - That is, now the determined scalars ^^^^^^^, ^^^^and ^^^^^^^, ^^^^per partition of the entire range of percentiles [0,100] are used for predicting the correction of the aggregated representative coarse image ^^^obtaining the coarse image at instant ^^co-registered and the spectral resolution adjusted. As it has been disclosed, when the images are multichannel, the correction of the aggregated representative coarse image ^^^is obtained for each channel recalculating scalars and ^^^^^^^, ^^^^ for the selected pixels at each partition. According to a plurality of embodiments that may be applied to former embodiment, These two combined conditions correspond to the specific case where images are multichannel images, each condition already disclosed above at the specific stage of the normalization method. When the partition of the interval [0,100] comprises a plurality (2 or more) of pairs of percentiles, for each pair of percentiles a set of selected pixels are identified. The union of the sets of pixels for all pairs is the complete image. Therefore, each linear regression determining are used for updating the set of pixels determined for the specific pair of percentiles ^^^^, ^^^^, that is, for a determined ^, the index identifying a partition of [0,100]. When using multichannel images, the set of pixels selected are according to a pre-specified channel or, according to a new channel generated by aggregating two or more of the available channels, and then such set of pixels is kept when processing all channels recomputing for each channel coefficients ^^^^^^^, ^^^^and ^^^^^^^, ^^^^of the linear regression. Therefore, when all pairs of percentiles are used, ^ = 1, … , ^, then all pixels of all image and all channels are also updated. According to an embodiment, aggregating two or more available channels is carried out by computing a linear combination of the set of values provided by each channel at the same pixel. According to a plurality of embodiments that may be applied to any former embodiment, the representative value is one of the following options: - the median value of the corresponding pixels values of the set of the ^ images or, - the mean value of the corresponding pixels values of the set of the ^ images or, - the maximum value of the corresponding pixels values of the set of the ^ images or, - a weighted value of the corresponding pixels values of the set of the ^ images or, - the pseudo-mean of the corresponding pixels values of the set of the ^ images. In this case, if the images are multichannel images, the representative value is computed for each channel using the values of the corresponding channel of the ^ images used in the calculation. As it has been previously disclosed, the preferred embodiment uses a pseudo-mean of the corresponding pixels values of the set of the ^ images since the new values are just a selection of the values already located on one of the images and do not require new calculations making the process more effective form the computational point of view. In any of the above embodiments the fine images and the coarse images correspond to the same region of the captured surface, have the same pose and are free of clouds. According to a further embodiment captured images are real images that are pre- processed by any of the following options: - by cropping one or more images so that the set of resulting images has the same size and each image corresponds to the same area of the captured surface; - by modifying the pose of one or more images so that the set of resulting images has the same pose; - by combining one or more captured images so that the set of resulting images has the same size and each image corresponds to the same area of the captured surface; - by filtering the presence of clouds; - by a combination of any previous one. According to an specific application captured images are captured by one or more satellites. In all embodiments, the method provides at least the generated image at ^^. The generated image may be provided by using a connecting bus, transmissions means or, being stored in storage means for a subsequent use. A second aspect of the invention is a computer system adapted to execute steps a) to g) according to any of the disclosed embodiments. A third aspect of the invention is a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of any of the disclosed embodiments. DESCRIPTION OF THE DRAWINGS These and other features and advantages of the invention will be seen more clearly from the following detailed description of a preferred embodiment provided only by way of illustrative and non-limiting example in reference to the attached drawings. Figure 1 This figure shows a scheme where a new fine image is generated from a plurality of images obtained from two different satellites. DETAILED DESCRIPTION OF THE INVENTION As will be appreciated by one skilled in the art, aspects of the present invention may be embodied as a system, method or computer program product. This embodiment describes a device for merging a set of images captured by cameras installed on board of two different satellites, with different resolutions, different poses, and with different spectral ranges. The device according to this embodiment has data storage and processing abilities, with the option to parallelize image processing and specific tasks. The device stores a set of images of an area of the earth that we will identify as ROI (region of interest) made during an interval of dates ^^^, ^^^, being ^^∈ ^^^, ^^^ the date for which the final image is wanted. The device of this example embodiment further has the capability to update the image database automatically for example by downloading images from public repositories. First of all, the device performs a preprocessing of the images so that they all correspond to the same region of the surface, have the same pose, all this for the set of fine and coarse images. In this preprocessing, the images are reprojected so that they all have the same pose, cropped if necessary and the clouds are filtered out, thus obtaining a set of new images which have the same resolution and extent. That is, at the end of the preprocessing two sets of images are obtained: fine resolution images ^^and coarse resolution images ^^where ^ denotes a common time base for all of them ^ = {^^, ^^, … , ^^}. For example, ^ = 6. According to this example, it will be seen how once the preprocessing of the images for the region of the surface is completed, the information from the fine resolution images will be used to obtain coarse resolution images that have compatible spectral information. This will be the normalization phase as described above. In a last phase the fine resolution image for the target date is generated by a local optimum prediction procedure to which a filter is optionally applied to remove noise. This procedure, thus identified in a general way, can be seen in Figure 1. In this figure 1, a rectangle with oblique sides represents input images. The embodiment uses satellite images from public repositories, in particular MODIS and Landsat-8 satellites, although the method is applicable to other combinations such as Sentinel 3 and Sentinel 2. In this embodiment the MODIS satellite provides images with a high temporal resolution, a daily image, but they are coarse resolution images, in this case 500m per pixel, and the Landsat-8 satellite provides fine resolution images, in particular 30 meters per pixel, although with a low temporal resolution, in this case one captured image every 16 days. In the example embodiment, enough images have been downloaded to cover the region of the surface by obtaining a single image by pre-processing that covers the region of the surface. In addition, a period of 48 days before ^^and 48 days after ^^has been downloaded, since Landsat-8 provides an image every 16 days. Landsat-8 images for the same ROI and different days allow to obtain a time series of fine images ^^or also explicitly denoted as ^^^with ^ = 1,2, … , ^. The same for MODIS images ^^^, ^^^, … , ^^^. The method requires a coarse, cloud-free image corresponding to the target date ^^to perform the prediction, in addition to coarse cloud-free images on the same dates for which fine images are available, in this embodiment example with a window of 15 days before and after, obtaining a coarse image for the target date, ^^^, and a series of images for the dates with Landsat-8 images, ^^^, … , ^^^. The method according to the example embodiment first performs a normalization process by transforming the thin images ^^and the coarse image of the target date ^^^such that they have similar spatial and spectral resolutions. Figure 1 shows at the left side and at the right side the normalization step. The set of captured fine images ^^are inputed in the scheme at the bottom of the left box and, the set of captured coarse images ^^are inputed at the upper right part of the right box. The set of captured fine images ^^are aggregated providing a set of coarse images ^^^. Normalization step provides a coarse-scale image target date ^^^^according to the steps already disclosed. The middle box in Figure 1 shows how the information captured by the cameras as well as the normalized images are combined. The goal of the local optimal prediction is to predict the fine resolution image ^^^^for the target date ^^by capturing the variation between the aggregated base period ^^^images and the corrected target date ^^^^image. With this objective, first, the optimal information is searched in the set of aggregated images ^^^and, the temporal changes are computed by linear regression between the optimal information and the target image ^^^^calculating, as described before. That is, the linear regression is calculated between the coarse images accumulated in a single image ^^^^^and the target coarse image ^^^^. The regression calculation allows determining the linear regression coefficients that will be used in the fine image. That is, the use of the linear regression allows extrapolating pixel-by-pixel the values of the fine image ^^^^from the fine images also aggregated by the use of the binary mask ^(^^, ^^)disaggretated from the binary mask ^(^^, ^^). The center of the middle box shows the plurality of channels showing the multichannel image generated from the binary mask ^(^^, ^^) and the images storing coefficients ^^(^^) and ^^(^^) that, after being represented as images are disaggregated for determining coefficients ^^^(^^), ^^^(^^) and ^(^^, ^^)respectively for calculating each pixel ^^of the fine image ^^^^at instant ^^as: wherein residual has also been calculated in order to increase the accuracy. New fine image ^^^^has the resolution of a captured image of Landstat-8.

Claims

CLAIMS 1.- Computer implemented method for generating an image of a surface at a predetermined time instant ^^, the method, executed by a computer system, comprising the steps: a) receiving a captured first fine image (^^^) at a time instant ^^of the surface, the first fine image (^^^) having a predetermined resolution; b) receiving a captured reference coarse image (^^^) at a time instant ^^of the surface wherein the reference coarse image (^^^) has a predetermined resolution lower than the resolution of the first fine image (^^^), the time instant ^^is different from the time instant ^^and, the reference coarse image (^^^) is co-registered with the first fine image (^^^); wherein steps a) and b) can be executed in any order and, c) generating a first coarse image (^^^^) by aggregating the first fine image (^^^) to the resolution of the reference coarse image (^^^); d) for each pixel ^^of the reference coarse image (^^^), o determining a vicinity region (V) comprising the pixel ^^and a predetermined set of pixels located at predetermined pattern around the pixel ^^; o determining a linear regression between the pixel values of an accumulated coarse image (^^^^^) and the pixels of the reference coarse image (^^^)wherein the accumulated coarse image (^^^^^) is generated by accumulating a set of aggregated captured fine images, the accumulated coarse image (^^^^^) being the first coarse image (^^^^) if there is only one captured fine image (^^^), the pixels used for calculating the linear regressions are the pixels of the vicinity region (V) of the accumulated coarse image (^^^^^) and the pixels of the vicinity region (V) of the first coarse image (^^^), in both cases the pixels of the vicinity region (V) are those located within the image and, linear coefficients ^^(^^), ^^(^^) and the residual ^(^^) are scalars, the three scalars determined by the linear regression; e) generating a first coarse coefficient image (^^^) storing ^^(^^)as pixels and a second coarse coefficient image (^^^) storing ^^(^^)as pixels;f) generating a first fine coefficient image (^^^) by disaggregating the first coarse coefficient image (^^^) and a second fine coefficient image (^^^) by disaggregating the second coarse coefficient image (^^^), the two generated fine coefficient images (^^^, ^^^) disaggregated according to the resolution of the first fine image (^^^); g) generating a fine image (^^^^) of the surface at time instant ^^by determining each pixel ^^value of the fine image (^^^^) as ^^^^(^^)= ^^^(^^)+ ^^^(^^)^^^^^(^^) wherein ^^^and ^^^are the corresponding disaggregated scalars ^^and ^^stored at pixel ^^of the corresponding coefficient image (^^^, ^^^) and ^^^^^(^^)is an accumulated fine image generated by accumulating the captured fine images, being the first fine image (^^^) if there is only one captured fine image. 2.- A method according to claim 1, wherein it further uses the calculated residual ^(^^)of pixel ^^resulting the linear regressionwherein - step e) further comprises generating a third coarse coefficient image (^^^) storing ^(^^)as pixels; - step f) further comprises generating a third fine coefficient image (^^^) by disaggregating the third coarse coefficient image (^^^), the generated fine coefficient image (^^^) disaggregated according to the resolution of the first fine image (^^^); and, - step g), when generating a fine image (^^^^) of the surface at time instant ^^by determining each pixel ^^value of the fine image (^^^^), the contribution for each pixel ^^value further comprises the valueof the corresponding disaggregated scalar ^(^^) stored at pixel ^^of the third coefficient image (^^^); that is,3.- A method according to claim 1 or 2, wherein the vicinity region (V) comprises a set of pixels located in a window ^^^^^=(2^^+ 1)^(2^^+ 1)around the pixel (^^) whereinand ^^are natural numbers of pixels. 4.- A method according to any of previous claims, wherein it further comprises: - in step a) capturing a plurality of ^ fine images (^^^,…, ^^^), the set of fine images atthe predetermined resolution of the first fine image (^^^), each fine image ^^^captured at a different time instant ^^; - in step c) generating a coarse image (^^^^) per fine image (^^^, ^ = 1, … , ^) by aggregating each fine image (^^^) to the resolution of the reference coarse image (^^^); - in step d) generating the accumulated coarse image (^^^^^) from the set of generated coarse images (^^^^) (^ = 1, … , ^), determining for each pixel ^^of the accumulated coarse image (^^^^^) and time instant ^^, ^ = 1, … , ^ of the base timewherein ^(^^, ^^)is a binary mask indicating which date ^^contains the optimal observation of pixel ^^and,denotes the correlation between the set of pixels of the vicinity region (V) of the reference coarse image (^^^) and the set of pixels of the vicinity region (V) of the said generated coarse image (^^^^), the correlation calculated in both cases with the pixels of the vicinity region (V) that are those located within the image; wherein the accumulated coarse is determined as ^^)^^^^(^^)- in step g), the accumulated fine image (^^^^^) is generated from the plurality of ^ fine images (^^^) (^ = 1, … , ^) according to the following sub-steps: o generating a fourth coarse coefficient image (^^^) storing the binary mask ^(^^, ^^), each channel of the multichannel image storing a time instant ^^; o generating a fourth fine coefficient multichannel image (^^^) by disaggregating the fourth coarse coefficient multichannel image (^^^); o determiningwherein ^(^^, ^^)is the disaggregated binary mask determined by the pixel values ^^of the disaggregated fine coefficient multichannel image (^^^). 5.- A method according to claim 4, wherein, the fourth coarse coefficient image (^^^) is a multichannel image storing the values of the binary mask ^(^^, ^^).6.- A method according to claim 4, wherein the fourth coarse coefficient image (^^^) is a single channel image wherein each pixel stores the index of the most representative image, the image having a ^(^^, ^^)= 1 value at said pixel. 7.- A method according to any of the previous claims, wherein generating a fine image from a coarse image by disaggregating said coarse image comprises: - disaggregating each pixel of the coarse image into a plurality of fine pixels of the fine image; - the pixel values of each plurality of pixels of the disaggregated pixels of the fine are representing the captured value at a predetermined location of the pixel and, its value is calculated by interpolating pixel values of the coarse image also representing the captured value at the same predetermined location of the pixel. 8.- A method according to the previous claim, wherein disaggregated pixels of the fine image generated from a pixel of the coarse image are different from disaggregated pixels of the fine image generated from a different pixel of the coarse image. 9.- A method according to any previous claims, wherein the predetermined location of the pixel is the center of the pixel. 10.- A method according to any of the previous claims, wherein at least the spectral resolution of the reference coarse image is adjusted by the following steps: - capturing a coarse image (^^^) at each time instant ^^^ = 1, .. , ^ of the surface wherein the coarse image (^^^) has a predetermined resolution lower than the resolution of the first fine image (^^^); and further executing by a computer system the steps: - generating a representative fine image ^^wherein each pixel value of the representative fine image ^^is calculated as a representative value of the corresponding pixels values of the set of the ^ fine images (^^^,…, ^^^); - generating a representative coarse image ^^wherein each pixel value of the representative coarse image ^^is calculated as the representative value of the corresponding pixels values of the set of the ^ captured coarse images (^^^,…,- aggregating the representative fine image ^^to a representative aggregated coarse image ^^^; - calculating the anomaly of ^^^as ^^^= ^^^− ^^, the anomaly of ^^^as ^^^=^^^− ^^and, the anomaly of ^^^^as ^^^= ^^^^− ^^, being ^^a pre-specified time instant among ^ = 1, .. , ^, preferably ^^= ^^wherein ^^is the time instant closest to ^^; - determining a partition of [0,100] by means of one or more pairs (^^^, ^^^), wherein ^^^< ^^^, for all ^ = 1, … , ^ being ^ the number of pairs of the partition and ⋃ ^ ^^^[^^^, ^^^] = [0,100]; - determining the variation between anomalies ^^^and ^^^by means of a linearwherein ^^^^^, ^^^^ is the residual and, the linear regression comprises the values of the set of pixels of the anomalies of ^^^^^^^, ^^^^and ^^^^^^^, ^^^^ranging between the percentiles (^^^, ^^^), that is, between pair of values at the same pixels, and wherein^^^^are two scalar values determined in the linear regression; - the anomaly at time instant ^^is estimated as ^^^^^^^, ^^^^ = ^^^^^^^, ^^^^ + ^^^^^^^, ^^^^^^^^^^^, ^^^^ wherein ^^^^^^^,, ^^^^are the values for the pixels used in the two linear regressions for each ^, wherein all pixels are updated for ^ = 1, … , ^; - the coarse image is adjusted at time instant ^^as:the adjustment being made for each pixel. 11.- A method according to previous claim, wherein the set of ^ fine images (^^, ^^, … , ^^), the set of captured coarse images (^^^, ^^^, … , ^^^) and the resulting representative fine image ^^and the representative coarse image ^^are multichannel images and, wherein the set of pixels selected as being those ranging between the percentiles (^^^, ^^^) are selected by using, or one single channel, a pre-defined channel, or using a new channel generated by aggregating two or more of the available channels, wherein the set of selected pixels is kept when computing the linear regression of ^^^^^^^, ^^^^ for a predetermined l value in the rest of channels but,coefficients and ^^^^^, ^^^^ residual being recomputed for every channel. 12.- A method according to previous claims, wherein a representative image from a setof ^ images is generated by determining each pixel value by calculating a representative value of the corresponding pixels values of the set of the ^ images, the representative value being one of the following options: - the median value of the corresponding pixels values of the set of the ^ images or, - the mean value of the corresponding pixels values of the set of the ^ images or, - the maximum value of the corresponding pixels values of the set of the ^ images or, - a weighted value of the corresponding pixels values of the set of the ^ images or, - the pseudo-mean of the corresponding pixels values of the set of the ^ images. 13.- A method according to any of previous claims, wherein captured images are pre- processed by any of the following options: - by cropping one or more images so that the set of resulting images has the same size and each image corresponds to the same area of the captured surface; - by modifying the pose of one or more images so that the set of resulting images has the same pose; - by combining one or more captured images so that the set of resulting images has the same size and each image corresponds to the same area of the captured surface; - by filtering the presence of clouds; - by a combination of any previous one. 14.- A method according to any of previous claims, wherein captured images are captured by one or more satellites. 15.- A computer system adapted to execute steps a) to g) according to any of claims 1 to 14. 16.- A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of any of claims 1 to 14.

Images