A curved surface guest sweat latent fingerprint extraction method and system

Abstract

The application discloses a curved-surface object sweat latent fingerprint extraction method and system, and relates to the field of sweat latent fingerprint extraction. The method comprises the following steps: obtaining a curved-surface sweat latent fingerprint image through a camera and three-dimensional data corresponding to the image; designing a curved-surface fingerprint flattening algorithm; correcting a curved-surface sweat latent fingerprint full-focus image caused by imaging perspective distortion of a curved-surface depth difference based on the three-dimensional data of the curved-surface sweat latent fingerprint image; and obtaining a flattened sweat latent fingerprint image. The flattened sweat latent fingerprint image can be obtained through only one curved-surface sweat latent fingerprint image obtained by the camera, without moving the camera position and without the aid of other rotating devices, so that the flattening problem of the sweat latent fingerprint image on the curved-surface object can be solved, and subsequent fingerprint comparison is facilitated.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of sweat latent fingerprint extraction, and particularly relates to a curved object sweat latent fingerprint extraction method and system. BACKGROUND

[0002] Fingerprint is the uneven line on the front surface of the human finger tip. Human fingerprint is unique, different and unchangeable for a lifetime, so it is also called the "second-generation ID card" of human body. Public security criminal investigation uses the characteristics of fingerprint to compare the fingerprint left in the scene with the existing fingerprint data to identify the identity information of the suspect and obtain important clues of the case. Fingerprint detection and identification have become an important technical means for extracting evidence in the process of criminal investigation.

[0003] The common fingerprint marks in criminal investigation scene can be mainly divided into visible fingerprint, sweat latent fingerprint and three-dimensional fingerprint, among which sweat latent fingerprint is the most common. However, sweat latent fingerprint is colorless and transparent, and is not easy to be found. The methods for detecting sweat latent fingerprint can be roughly divided into physical detection, chemical detection and optical detection. Physical detection uses the sweat adsorption to adsorb other substances on the fingerprint line through iodine fumigation, 502 glue fumigation and other methods for color development; chemical detection uses the chemical reaction between the substances in sweat latent fingerprint and chemical reagents for fingerprint development. The methods for detecting sweat latent fingerprint by physical detection and chemical detection will damage the original fingerprint in the scene and affect the secondary evidence of the fingerprint clue. The method for detecting sweat latent fingerprint by optical detection will not damage the fingerprint, and can detect the fingerprint twice, so it is widely used in the field of public security criminal investigation.

[0004] Ultraviolet reflection photography is one of the methods for optically detecting sweat latent fingerprint. This method uses the different absorption and reflection abilities of fingerprint lines and background to ultraviolet light to record the brightness distribution of fingerprint reflected ultraviolet light, and is a sweat latent fingerprint non-destructive detection technology with good development effect and repeated development. In addition, studies have shown that ultraviolet short-wave lamp irradiation within ten minutes does not affect the subsequent DNA detection of sweat latent fingerprint. At present, the criminal evidence camera based on ultraviolet reflection direct imaging can quickly extract sweat latent fingerprint on the plane and plays an important role in criminal investigation. For sweat latent fingerprint on curved objects, such as fingerprint left on curved surfaces such as handles, railings and cups, because of the depth difference of curved surface, ordinary criminal investigation camera cannot get clear fingerprint images at all places at one time, and the local images of fingerprint at different depths have different degrees of perspective distortion, so the obtained fingerprint images cannot be directly used for subsequent fingerprint comparison, and the curved fingerprint images need to be flattened.

[0005] For the problem of sweat latent fingerprint extraction of curved object, some researchers take multi-angle photos of the fingerprints distributed on the curved object, and then obtain the complete fingerprint image through image stitching; some researchers use short-wave ultraviolet reflection photography method and mechanical and electrical driving rotating stage linkage to rotate the curved surface for multiple shooting, stitch the series of partial images of the curved surface fingerprints obtained by shooting, and complete the extraction of the fingerprints on the surface of the curved object. Whether rotating the camera or rotating the shooting object, the flattened image of the entire fingerprint on the curved surface cannot be obtained by one-time imaging, and the extraction efficiency is low. Therefore, it is of great significance to quickly and non-destructively extract the flattened sweat latent fingerprint of the curved object for the criminal investigation process. SUMMARY

[0006] The technical problem to be solved by the present application is that in the prior art, for the problem of sweat latent fingerprint extraction of curved object, whether rotating the camera or rotating the shooting object, the flattened image of the entire fingerprint on the curved surface cannot be obtained by one-time imaging, and the extraction efficiency is low.

[0007] To solve the above technical problems, the present application adopts the following technical solutions:

[0008] A sweat latent fingerprint extraction method of curved object, for the sweat latent fingerprint on the curved object, based on the curved sweat latent fingerprint image obtained by the camera and the three-dimensional data corresponding to the image, the following steps are performed to obtain the flattened complete sweat latent fingerprint image:

[0009] Step A1: based on the curved sweat latent fingerprint image and the three-dimensional data corresponding to the image, the curved sweat latent fingerprint image is divided into regions with overlap;

[0010] Step A2: local texture correction is performed for each region to obtain a uniform sampling correction texture image corresponding to each region;

[0011] Step A3: image stitching is performed on the uniform sampling correction texture image corresponding to each region to obtain the flattened complete sweat latent fingerprint image.

[0012] As a preferred technical solution of the present application, in step A1, the following steps are included:

[0013] Step A1.1: for the three-dimensional data of the curved sweat latent fingerprint image, the k-means clustering algorithm is used to divide the three-dimensional data points corresponding to the curved sweat latent fingerprint image, and the curved sweat latent fingerprint image is segmented into k non-overlapping small regions according to the division result of the three-dimensional data points, and the value of k is determined by using the elbow method on the k-means clustering objective function;

[0014] Step A1.2: for each sub-region, an inflation operation in morphological image processing is used to make adjacent sub-regions overlap with each other, to obtain each region with overlap, and the overlap rate is determined by a preset inflation operator radius.

[0015] As a preferred technical solution of the present application, in step A2, the following steps are included:

[0016] Step A2.1: for the three-dimensional data corresponding to each region, a least square method is used to perform plane fitting on the three-dimensional data of each region respectively, to obtain the normal vector of each fitting plane corresponding to each region;

[0017] Step A2.2: the direction indicated by the normal vector of each fitting plane is taken as the orthographic view angle direction corresponding to each region respectively, and perspective correction is performed on each region based on the normal vector of each fitting plane, to obtain the three-dimensional coordinates of each pixel point of each region under the orthographic view angle as the three-dimensional homogeneous coordinates of each pixel point of each region;

[0018] Step A2.3: by dividing by the same preset scale factor, the three-dimensional homogeneous coordinates of each pixel point of each region are converted into two-dimensional non-homogeneous coordinates, to obtain the non-homogeneous coordinates of each pixel point of each region as the two-dimensional position of each pixel point of each region, and then to obtain the corrected texture image of each region with non-uniform sampling;

[0019] Step A2.4: for each region, the corrected texture image with non-uniform sampling is subjected to uniform grid interpolation, to obtain the uniform sampling corrected texture image corresponding to each region.

[0020] As a preferred technical solution of the present application, in step A3, for the uniform sampling corrected texture image corresponding to each region, a region-based image stitching method is used, normalized cross-correlation is used as a similarity measure to calculate the relative offset between adjacent uniform sampling corrected texture images, and then the uniform sampling corrected texture image corresponding to each region is stitched into a flattened complete latent fingerprint image according to the relative offset.

[0021] As a preferred technical solution of the present application, in step A2.2, the perspective correction on each region based on the normal vector of each fitting plane is specifically a camera coordinate system rotation according to the orthographic view angle direction corresponding to each region, and the three-dimensional coordinates of each pixel point of each region image under the orthographic view angle are obtained after the camera coordinate system rotation of each region image; for the i-th region, the camera coordinate system rotation formula is as follows:

[0022] x′=KRK -1 x

[0023] wherein,

[0024] In the formula, x represents the three-dimensional coordinates of any pixel point on the i-th region in the camera coordinate system, x' represents the three-dimensional coordinates of the pixel point after rotation of the camera coordinate system; K represents the camera intrinsic matrix; R represents the camera coordinate system rotation matrix; n i represents the fitting plane normal vector of the i-th region image;(i A , j A , k A ) represents the camera coordinate system;(i B , j B , k B ) represents the coordinate system after rotation of the camera coordinate system.

[0025] A system based on the curved surface object sweat latent fingerprint extraction method, comprising a region division module, a region correction module, a splicing module,

[0026] The region division unit is used to divide the curved surface sweat latent fingerprint image into regions with mutual overlap based on the curved surface sweat latent fingerprint image and the three-dimensional data corresponding to the image;

[0027] The region correction unit is used to perform local texture correction on each region respectively to obtain the uniform sampling correction texture image corresponding to each region respectively;

[0028] The splicing unit is used to perform image splicing on the uniform sampling correction texture image corresponding to each region respectively to obtain the flattened complete sweat latent fingerprint image.

[0029] A terminal applied to the curved surface object sweat latent fingerprint extraction method, comprising a memory and a processor, which are mutually communicatively connected, and the memory stores computer instructions, and the processor executes the computer instructions to perform the curved surface object sweat latent fingerprint extraction method.

[0030] The curved surface object sweat latent fingerprint extraction method and system provided by the application can solve the problem of flattening the sweat latent fingerprint image on the curved surface object, and facilitate subsequent fingerprint comparison. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 is the light path principle diagram of the ultraviolet focusing type light field camera in the embodiment of the application;

[0032] Figure 2 is a flow chart of a curved surface fingerprint flattening algorithm in an embodiment of the present application;

[0033] Figure 3 is a schematic diagram of local texture correction in an embodiment of the present application. DETAILED DESCRIPTION

[0034] The present application will be further described below with reference to the drawings. The following embodiments enable a person skilled in the art to more fully understand the present application, but do not limit the present application in any way.

[0035] A curved surface latent sweat fingerprint extraction method, for a curved surface latent sweat fingerprint on a curved surface object, based on a curved surface latent sweat fingerprint image obtained through a camera and three-dimensional data corresponding to the image, the following steps are performed to obtain a flattened complete latent sweat fingerprint image.

[0036] In this embodiment, for a curved surface latent sweat fingerprint image and three-dimensional data corresponding to the image, an ultraviolet focusing type light field camera as shown in Figure 1 is used to obtain, specifically, a curved surface latent sweat fingerprint full-focus image obtained through an image reconstruction algorithm of the ultraviolet focusing type light field camera is used as the curved surface latent sweat fingerprint image, and three-dimensional data corresponding to the curved surface latent sweat fingerprint image obtained through a depth reconstruction algorithm of the ultraviolet focusing type light field camera. The ultraviolet focusing type light field camera adopts a focusing type light field camera structure, that is, each structural device in the focusing type light field camera is transformed into a corresponding device suitable for ultraviolet imaging, and is composed of an ultraviolet light source, an ultraviolet main lens, an ultraviolet microlens array, an ultraviolet band-pass filter, and an ultraviolet enhanced image sensor. The ultraviolet light source surrounds the ultraviolet main lens and irradiates the curved surface object with light.

[0037] The light of the ultraviolet light source irradiates the latent sweat fingerprint on the curved surface object, and the ultraviolet light beam reflected by the fingerprint is collected by the ultraviolet main lens and first virtually imaged to a virtual image plane. Each microlens of the ultraviolet microlens array focuses the local part of the first virtual image plane twice, and the converging light is filtered by the ultraviolet band-pass filter to remove background light of other wavebands, and then converges to the ultraviolet enhanced image sensor for imaging, thereby completing ultraviolet light field collection of the curved surface latent sweat fingerprint. The collected light field information is encoded in the image obtained by the ultraviolet enhanced image sensor. The image is composed of microlens sub-images on the microlens array that participate in the secondary imaging. Since the microlens aperture is small, each sub-image can clearly image the local part of the curved surface fingerprint within the depth of field of the microlens. Alternatively, other ultraviolet imaging and three-dimensional measurement techniques can be used to obtain the curved surface latent sweat fingerprint image and the three-dimensional data corresponding to the image.

[0038] The traditional ultraviolet criminal investigation camera is improved in the embodiment. In order to facilitate the collection of the curved sweat latent fingerprint, the focusing light field camera is introduced into the field of ultraviolet fingerprint imaging. Without moving the camera or the position of the fingerprint carrier, the light field image obtained by single shooting can obtain a clear curved sweat latent fingerprint image everywhere after reconstruction, and the extraction efficiency of the curved sweat latent fingerprint is improved.

[0039] The image reconstruction algorithm and the depth reconstruction algorithm of the ultraviolet focusing light field camera use the existing classical image reconstruction algorithm and depth reconstruction algorithm of the focusing light field camera. The image reconstruction algorithm uses the characteristic that the imaging content of each microlens sub-image of the focusing light field camera is overlapped, uses the normalized cross-correlation to calculate the relative offset between adjacent microlens sub-images, and sequentially splices all the microlens sub-images into a full-focus image of a full field of view based on the obtained relative offset. The depth reconstruction algorithm is based on the binocular disparity principle, calculates the depth of a point on the target based on the relative positions of different microlens sub-images when the point is imaged, realizes the depth estimation of the curved object, and further obtains the three-dimensional data of the curved sweat latent fingerprint image.

[0040] In the embodiment, the above-mentioned ultraviolet focusing light field camera is used to collect, reconstruct and reconstruct the three-dimensional data of the sweat latent fingerprint on the curved object, to obtain the curved sweat latent fingerprint image and its three-dimensional data, and then the curved fingerprint flattening algorithm described in steps A1-A3 is executed, as shown in Figure 2 The curved fingerprint flattening algorithm is based on the three-dimensional data of the curved fingerprint to flatten the full-focus image of the curved fingerprint. The algorithm first divides the full-focus image of the curved fingerprint into regions with overlapping each other using a clustering algorithm, then performs plane fitting on each region, corrects the perspective distortion of each region based on the normal vector of the fitting plane, and finally splices the corrected texture local into a flattened fingerprint image.

[0041] Step A1: based on the curved sweat latent fingerprint image and the three-dimensional data corresponding to the image, the three-dimensional data being three-dimensional coordinate data corresponding to the image, the curved sweat latent fingerprint image is divided into regions with overlapping each other, and the curved sweat latent fingerprint image is a two-dimensional image obtained by a camera.

[0042] In the step A1, the following steps are included:

[0043] Step A1.1: for the three-dimensional data of the curved sweat latent fingerprint image, the k-means clustering algorithm is used to divide the three-dimensional data points corresponding to the curved sweat latent fingerprint image, and the curved sweat latent fingerprint image is segmented into k non-overlapping small regions according to the division result of the three-dimensional data points, and the value of k is determined by using the elbow method on the k-means clustering objective function; the curved segmentation result is to divide the whole curved distortion image into k small regions with overlapping each other.

[0044] Step A1.2: For each sub-region, an inflation operation in morphological image processing is used to make adjacent sub-regions overlap with each other, to obtain regions that overlap with each other, and the overlap rate is determined by a preset inflation operator radius. This step is mainly to facilitate the implementation of algorithm image stitching; in this embodiment, the inflation operator radius is set to half of the region diameter, to ensure a 50% region overlap rate, and to obtain a better image stitching effect.

[0045] Step A2: Local texture correction is performed on each region, to obtain a uniform sampling corrected texture image corresponding to each region.

[0046] In step A2, the following steps are included:

[0047] Step A2.1: For the three-dimensional data corresponding to each region, a least squares method is used to perform plane fitting on the three-dimensional data of each region, to obtain a normal vector of each fitting plane corresponding to each region.

[0048] Step A2.2: The direction indicated by the normal vector of each fitting plane is taken as the front view perspective direction corresponding to each region, that is, the perspective that conforms to the normal vector direction is the front view perspective, and perspective correction is performed on each region based on the normal vector of each fitting plane, to obtain the three-dimensional coordinates of each pixel point of each region under the front view perspective; as the three-dimensional homogeneous coordinates of each pixel point of each region.

[0049] In step A2.2, perspective correction is performed on each region based on the normal vector of each fitting plane, specifically, a camera coordinate system rotation is performed according to the front view perspective direction corresponding to each region, and the three-dimensional coordinates of each pixel point of each region image under the front view perspective are obtained after the camera coordinate system rotation of each region image; for the i-th region, the camera coordinate system rotation formula is as follows:

[0050] x′=KRK -1 x

[0051] Wherein,

[0052] In the formula, x represents the three-dimensional coordinates of any pixel point on the i-th region under the camera coordinate system, x' represents the three-dimensional coordinates of the pixel point after the camera coordinate system rotation; K represents the camera intrinsic parameter matrix; R represents the camera coordinate system rotation matrix; n i represents the normal vector of the fitting plane of the i-th region image; (i A , j A , k A ) represents the camera coordinate system; (i B , j B , k B ) represents the coordinate system after the camera coordinate system rotation.

[0053] The three-dimensional coordinates of each pixel in each region of the image under the frontal view are the three-dimensional homogeneous coordinates of the two-dimensional pixel coordinates of each region. They need to be converted into the corresponding two-dimensional non-homogeneous coordinates to indicate the corrected two-dimensional position of the pixel.

[0054] Step A2.3: By dividing by the same preset scale factor, the 3D homogeneous coordinates of each pixel in each region are transformed into 2D non-homogeneous coordinates, obtaining the non-homogeneous coordinates of each pixel in each region. These non-homogeneous coordinates serve as the 2D positions of each pixel in each region, thus obtaining the corrected texture image of each region after non-uniform sampling. The scale factor is determined by the average depth of the surface. Specifically, the transformation of the 3D homogeneous coordinates of each pixel in each region into 2D non-homogeneous coordinates involves dividing the 3D coordinates of each pixel in each region by the preset scale factor to obtain new 3D coordinates. The first two dimensions of the new 3D coordinates are then taken as the 2D non-homogeneous coordinates, i.e., the non-homogeneous coordinates of each pixel in each region. In the 3D coordinate system, the first dimension corresponds to the vertical direction, the second dimension corresponds to the horizontal direction, and the third dimension corresponds to the depth direction.

[0055] Step A2.4: Perform uniform grid interpolation on the corrected texture images of each region that are non-uniformly sampled to obtain the uniformly sampled corrected texture images for each region.

[0056] Step A2 is for local texture correction of each region; firstly, the least squares method is used to perform planar fitting on the 3D data corresponding to each region, and the normal vector of each fitted plane is calculated. Based on the normal vector of the fitted plane, the perspective distortion texture image of each region is corrected to a texture image of the same scale under the corresponding frontal view. Figure 3 As shown, the small region π i The fitting plane normal vector n i The indicated direction is the frontal viewing direction of the corresponding area. Perspective correction is performed on the texture of this area, which involves rotating the camera coordinate system along this direction and calculating the three-dimensional coordinates of each pixel in the area after the camera rotation. Based on the pinhole camera imaging model, the region π i The three-dimensional coordinates x of the image of the previous object point X on the original imaging plane 1, and the three-dimensional coordinates x′ of the image after rotation in the camera coordinate system, can be expressed as:

[0057] x′=KRK -1 x

[0058] Where K is the camera intrinsic parameter matrix, and R is the camera coordinate system rotation matrix. Let the original camera coordinate system be (i A j A k A After rotation, the camera coordinate system is (i B j B k B ), n i For small region πi the normal vector of the fitted plane, and

[0059]

[0060] where · denotes the vector dot product and × denotes the vector cross product. The rotation matrix R is

[0061]

[0062] Since each sub-region is corrected separately, the problem of camera center translation does not need to be considered, Figure 3 The camera centers C and C' before and after rotation are actually coincident, i.e., the origins of the coordinate systems before and after rotation are actually coincident. To facilitate subsequent stitching, the scale of the texture after correction of each region needs to be adjusted to be uniform. When converting x' to two-dimensional non-homogeneous coordinates, the same scale factor can be uniformly divided. The same scale factor means that the distance from the camera after rotation to each sub-region is the same. The local corrected texture image obtained after the above correction and scale adjustment has a non-uniform distribution of coordinate points. The last step of local texture correction is to perform uniform grid interpolation on the corrected texture image with non-uniformly distributed coordinates.

[0063] Step A3: For each region, the corresponding uniformly sampled corrected texture image is subjected to image stitching to obtain a flattened complete sweat latent fingerprint image.

[0064] In the step A3, for each region, the corresponding uniformly sampled corrected texture image is subjected to image stitching using a region-based image stitching method. Normalized cross-correlation is used as a similarity measure to calculate the relative offset between adjacent uniformly sampled corrected texture images. Then, the relative offset is used to stitch the corresponding uniformly sampled corrected texture images of each region into a flattened complete sweat latent fingerprint image.

[0065] Based on the above method, the embodiment also provides a system based on the sweat latent fingerprint extraction method of the curved object, which comprises a region division module, a region correction module, a stitching module,

[0066] The region division unit is configured to divide the curved sweat latent fingerprint image into regions that overlap with each other based on the curved sweat latent fingerprint image and three-dimensional data of the image.

[0067] The region correction unit is configured to perform local texture correction on each region to obtain a uniformly sampled corrected texture image corresponding to each region.

[0068] The stitching unit is configured to perform image stitching on the uniformly sampled corrected texture image corresponding to each region to obtain a flattened complete sweat latent fingerprint image.

[0069] The embodiment also provides a terminal applied to the curved object sweat latent fingerprint extraction method, including a memory and a processor, which are in communication connection with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the curved object sweat latent fingerprint extraction method.

[0070] The present application designs a curved object sweat latent fingerprint extraction method and system, for the sweat latent fingerprints distributed on the curved object, without moving the position and without the help of other rotating devices, the sweat latent fingerprints are collected by an ultraviolet focusing light field camera, the obtained light field image is decoded by a focusing light field camera image reconstruction algorithm to obtain a clear curved sweat latent fingerprint full-focus image everywhere, and three-dimensional data of the curved fingerprint is obtained by a depth reconstruction algorithm of the focusing light field camera. A curved fingerprint flattening algorithm is designed, based on the three-dimensional data of the curved sweat latent fingerprint, the curved sweat latent fingerprint full-focus image which is distorted by the imaging perspective of the curved depth difference is corrected to obtain a flattened sweat latent fingerprint image. The collection efficiency of the curved sweat latent fingerprint on the curved object in the criminal investigation scene can be improved, the problem of flattening the curved sweat latent fingerprint is solved, and the subsequent fingerprint comparison is facilitated.

[0071] The above is only the preferred embodiment of the present application, but does not limit the patent range of the present application, although the present application is described in detail with reference to the foregoing embodiment, for those skilled in the art, the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced equivalently. Any equivalent structure made by using the content of the present application specification and drawings, directly or indirectly used in other related technical fields, is also within the patent protection scope of the present application.

Claims

1. A method for latent fingerprint extraction from curved surfaces of objects, characterized by, For the sweat latent fingerprints on the curved surface object, based on the curved surface sweat latent fingerprint image obtained by the camera and the three-dimensional data corresponding to the image, the following steps are performed to obtain the flattened complete sweat latent fingerprint image: Step A1: based on the curved surface sweat latent fingerprint image and the three-dimensional data corresponding to the image, the curved surface sweat latent fingerprint image is divided into regions with overlap with each other; Step A2: local texture correction is performed for each region respectively to obtain the uniform sampling corrected texture image corresponding to each region respectively; Step A3: for the uniform sampling corrected texture image corresponding to each region, image stitching is performed to obtain the flattened complete sweat latent fingerprint image; Step A1 includes the following steps: Step A1.1: for the three-dimensional data of the curved surface sweat latent fingerprint image, the k-means clustering algorithm is used to divide the three-dimensional data points corresponding to the curved surface sweat latent fingerprint image, and the curved surface sweat latent fingerprint image is segmented into k non-overlapping small regions according to the division result of the three-dimensional data points, and the value of k is determined by using the elbow method on the k-means clustering objective function; Step A1.2: for each small region, an inflation operation in morphological image processing is used to make the adjacent small regions overlap with each other, and the regions with overlap with each other are obtained, and the overlap rate is determined by a preset inflation operator radius; Step A2 includes the following steps: Step A2.1: for the three-dimensional data corresponding to each region, least squares method is used to fit a plane for the three-dimensional data of each region respectively, and then the normal vector of each fitting plane corresponding to each region is obtained; Step A2.2: the direction indicated by the normal vector of each fitting plane is taken as the orthographic view angle direction corresponding to each region respectively, perspective correction is performed on each region based on the normal vector of each fitting plane, and the three-dimensional coordinates of each pixel point of each region under the orthographic view angle are obtained as the three-dimensional homogeneous coordinates of each pixel point of each region; Step A2.3: by dividing by the same preset scale factor, the three-dimensional homogeneous coordinates of each pixel point of each region are converted into two-dimensional non-homogeneous coordinates, and the non-homogeneous coordinates of each pixel point of each region are obtained as the two-dimensional position of each pixel point of each region, and then the non-uniformly sampled corrected texture image of each region is obtained; Step A2.4: for each non-uniformly sampled corrected texture image of each region, uniform grid interpolation is performed to obtain the uniform sampling corrected texture image corresponding to each region.

2. The method as claimed in claim 1, wherein, In the step A3, for the uniform sampling corrected texture image corresponding to each region, a region-based image stitching method is used, the normalized cross-correlation is used as the similarity measure to calculate the relative offset between adjacent uniform sampling corrected texture images, and then the uniform sampling corrected texture image corresponding to each region is stitched into the flattened complete sweat latent fingerprint image according to the relative offset.

3. The method as claimed in claim 1, wherein the curved surface object is a finger. In the step A2.2, the perspective correction is performed on each region based on the normal vector of each fitting plane, specifically, the camera coordinate system rotation is performed according to the orthographic view angle direction corresponding to each region, and the three-dimensional coordinates of each pixel point of each region image under the orthographic view angle are obtained after the camera coordinate system rotation of each region image; for the i-th region, the camera coordinate system rotation formula is as follows: ； wherein , wherein, represents the three-dimensional coordinates of any pixel point on the i-th region in the camera coordinate system, represents the three-dimensional coordinates of the pixel point after rotation in the camera coordinate system; represents the intrinsic matrix of the camera; represents the rotation matrix of the camera coordinate system; represents the fitting plane normal vector of the i-th region image; represents the camera coordinate system; represents the coordinate system after rotation of the camera coordinate system.

4. A system for latent fingerprint extraction based on the method of any one of claims 1 to 3, characterized in that, The system includes a region division module, a region correction module, and a stitching module, The region division unit is configured to divide the curved latent sweat fingerprint image into regions with overlaps based on the curved latent sweat fingerprint image and three-dimensional data corresponding to the image. The region correction unit is configured to perform local texture correction on each region respectively to obtain a uniform sampling corrected texture image corresponding to each region respectively. The splicing unit is configured to perform image splicing on the uniform sampling corrected texture images corresponding to each region respectively to obtain a flattened complete latent sweat fingerprint image.

5. A terminal for applying the method for latent fingerprint extraction from the curved surface object, characterized in that, A device includes a memory and a processor in communication with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method of any one of claims 1-3.

Images