A method for suppressing stray light of a curtain image under multi-light-source projection

CN122698730APending Publication Date: 2026-09-04CHONGQING YINGWEI AUDIOVISUAL EQUIPMENT CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610833996.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-06-10
Publication Date
2026-09-04

AI Technical Summary

Technical Problem

[0003]但是现有方案处理多投影杂光时,大多只做重叠区的亮度羽化融合,并未在融合前剥离各台投影间的互扰成分,导致串扰引起的色偏和亮度脏污被一并混合进最终画面,融合带的均匀性难以根除

Benefits of technology

[0056] 1. In this invention, by utilizing a pre-established stray light crosstalk weight table and a high-speed rotating synchronous acquisition link, the instantaneous individual images of four projectors can be completely captured in each round of four-frame shooting. Through the joint inverse operation of the temporal mixing matrix and the spatial crosstalk matrix, residual light caused by device switching trailing, shutter shake, and screen mutual reflection is removed in one go, resulting in a pure component free of stray light. This deterministically decomposes the multi-source brightness originally superimposed in each frame into independent channels, ensuring that the input for subsequent pre-compensation curve calculations is not contaminated by crosstalk, preventing crosstalk errors from being further amplified during the pre-compensation stage, increasing the upper limit of the signal-to-noise ratio of the entire correction link, and improving the overall suppression effect.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122698730A_ABST
    Figure CN122698730A_ABST
Patent Text Reader

Abstract

The application discloses a kind of multi-light source projection under curtain image stray light suppression processing method.In the present application, with the aid of pre-established stray light crosstalk weight table and high-speed wheel launch synchronous acquisition link, four frames of shooting in each round can capture four instantaneous single-shot pictures projected completely, and through the joint inverse operation of time mixing matrix and space crosstalk matrix, residual one-time stripping caused by device switch tailing, shutter shake light penetration and curtain mutual reflection is obtained, which does not contain stray light pure component. Thus, the multi-source brightness originally mixed in each frame is determined to be decomposed into independent channels, so that the calculation input of subsequent pre-compensation curve is not contaminated by crosstalk, the upper limit of signal-to-noise ratio of the whole correction link is improved, the overall suppression processing effect is improved, the acquisition of pure component makes the pre-compensation texture can be baked into lookup table texture resident memory at one time, without online iterative solution, which significantly reduces the computational overhead of real-time processing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically a method for suppressing stray light in screen images under multi-source projection. Background Technology

[0002] Multi-projection screen image stray light processing refers to the collective term for optical suppression and digital correction techniques employed to address the problems of unwanted stray light overflow and contrast attenuation caused by optical path overlap, screen diffuse reflection, and ambient light interference when multiple projectors are spliced ​​and blended for display. In multi-projection systems, the superposition of light intensity in the edge overlap area easily produces bright bands, and non-ideal scattering from the projection screen often causes halos and color pollution, severely affecting the uniformity and immersion of the image. This processing technology mainly covers two dimensions: hardware optical avoidance and software algorithm elimination. At the hardware level, it uses high-gain directional reflective screens, anti-glare screens, and aperture masks to physically suppress large-angle scattering and ambient light reflection. At the software level, it relies on optomechanical calibration and photometric measurement, and uses digital algorithms such as geometric mapping, alpha channel feathering blending, stray light dot matrix sampling, and deconvolution compensation to accurately calculate and cancel the brightness and color interference of stray light on effective pixels. This processing technology is a key link in ensuring that multi-projection blending systems achieve high dynamic range, seamless splicing, and high color reproduction, and is widely used in high-end visual fields such as virtual simulation, immersive CAVE systems, and large command and control centers.

[0003] However, existing solutions for handling stray light from multiple projections mostly only perform brightness feathering and blending in the overlapping areas, without removing the mutual interference components between the projectors before blending. This results in color shifts and brightness contamination caused by crosstalk being mixed into the final image, making it difficult to eliminate the uniformity of the blended band. At the same time, ambient light compensation generally adopts a global light sensing and overall brightening and darkening approach, which is disconnected from pixel-by-pixel crosstalk correction. This causes the input signal of the pre-compensation stage to contain crosslight that drifts with the aging of the projector or changes in the environment. This forces the online system to repeatedly iterate and estimate or rely on manual experience to repeatedly adjust parameters, resulting in poor real-time performance and insufficient long-term stability. Summary of the Invention

[0004] The purpose of this invention is to provide a method for suppressing stray light in screen images under multi-source projection in order to solve the problems mentioned above.

[0005] The technical solution adopted in this invention is as follows: A method for suppressing stray light in screen images under multi-source projection, comprising the following steps:

[0006] S1: Turn on the projector and project a full white screen. Use a calibration camera to photograph the screen to obtain the light intensity attenuation distribution of each projector at each pixel position. Organize these distribution data into a lighting base map.

[0007] S2: Based on the illumination substrate atlas obtained in step S1, calculate the energy superposition ratio of multiple light sources simultaneously illuminating each position on the screen, and summarize it into a stray light crosstalk weight table. This table will be provided to S3 as a priori reference for separation.

[0008] S3: Put all projectors into high-speed rotation mode, flashing test light alternately at a frequency much higher than the display frame rate. The camera shutter is strictly synchronized with the light emission window of each projector through hard triggering, directly capturing the instantaneous single-image of each projector. After capturing, the crosstalk weight table of S2 is called to perform secondary removal of inter-frame light leakage residue, resulting in a pure component with almost no stray light.

[0009] S4: Compare the pure component obtained in S3 with the target ideal image point by point, calculate the pre-compensation curve in the form of an inverse function, and bake the curve into a compensation texture that can be directly sampled for subsequent real-time rendering.

[0010] S5: In actual playback, each frame of the image first pre-distorts the pixels according to the compensation texture of S4, and at the same time receives the readings of the ambient light probe near the screen, and the overall bias compensation texture response substrate is used to track and counteract the dynamic changes of ambient stray light.

[0011] S6: Receives the pre-distorted multi-channel images from S5, performs brightness feathering attenuation in the projection overlap area to make the blending band smooth and seamless, and synchronously outputs them to the projector array to complete the full display after stray light suppression.

[0012] In a preferred embodiment, in step S1, after the multi-source projection system is set up, the color temperature of all projectors is first unified to around 6500K through white balance calibration, and the brightness non-uniformity is controlled within ±3%. All ambient light sources are turned off, and a standard white board is attached to the screen surface for reference calibration. The camera is fixed at the viewer's viewpoint, with the angle between the optical axis and the screen normal less than 2 degrees. During shooting, a completely black frame is first captured as the dark current reference, and then the projectors are turned on to project a completely white image. Four frames are taken from each projector separately, and the median is averaged to suppress random noise. The exposure time is set so that the histogram peak falls around 70% of the full-well capacity. Each white field image after dark field subtraction is directly recorded as the illumination base map corresponding to that projector. The spatial resolution of all layers is consistent with the final display resolution, without interpolation scaling. The pixel value unit is the digital quantization value output by the camera, which is directly mapped to the relative light intensity at that location, forming the illumination base map set required for subsequent steps.

[0013] In a preferred embodiment, in step S2, to obtain the stray light crosstalk weights pixel by pixel, four sets of brightness readings corresponding to each pixel position are first extracted from the illumination substrate atlas transmitted from S1, corresponding to the measurement values ​​when the four projectors are lit individually. Utilizing the diffuse reflection symmetry of the screen, the brightness disturbance of a single projector at a certain pixel is attributed to the local illuminance gradient formed by other projected light spots near that pixel. The contribution ratio of direct light from each projector to secondary scattering from the screen is fixed using the idea of ​​light tracing, thereby obtaining a 4×4 crosstalk weight matrix. Finally, a stray light crosstalk weight table with the same resolution as the screen is compiled and directly passed to S3 as a priori matrix for separation.

[0014] In a preferred embodiment, in step S2, the formula for calculating the elements of the crosstalk weight matrix is:

[0015]

[0016] In the formula:

[0017] Indices j and i refer to the projector affected by crosstalk and the projector that is the source of stray light, respectively.

[0018] Index k is the loop variable used to iterate through all four projectors when summing the denominators. The values ​​of the three indices are all between 1 and 4.

[0019] This represents the brightness value recorded by the S1 illumination substrate atlas at the current pixel when only projection j is lit, and the unit is the number of digit levels output by the camera. The same meaning and corresponding to the projector pointed to by the summation variable k.

[0020] Let be the local bidirectional reflectance of the screen at this pixel, describing the light scattering efficiency from the projection j outgoing direction to the projection i receiving direction. This value is obtained by considering the local bidirectional reflectance of the pixel's neighborhood in the S1 illumination substrate map. Spatial gradient and The spatial gradient is calculated jointly, reflecting the directional scattering characteristics of the screen microstructure for incident light. The denominator contains... Similarly, this applies to each projector k.

[0021] This is the correction factor for the solid angle subtended by the projection lens i for that pixel. It is calculated based on the principal exit direction of projection j. Its value is obtained in advance through geometric calibration based on the projector's optical center position, the spatial coordinates of the pixel on the screen, and the screen surface normal vector. The denominator contains... Similarly.

[0022] Summation term in denominator The weighted sum of the scattered light flux of all projections along the receiving direction of projection i is used as the normalization reference. This normalization design ensures that for any fixed receiving projection i, all its row directions are normalized. The sum of all values ​​is always equal to 1, reflecting the total stray light contribution distributed to each source projection according to energy proportion, and the matrix naturally maintains the energy closure property that the row sum is 1.

[0023] In a preferred embodiment, during step S3, when establishing the synchronous acquisition link, four identical DLP projectors are used, and their LED driver boards are connected to a 240Hz square wave timing signal generated by an FPGA. The four enable signals output by the FPGA are sequentially offset by 90 degrees, causing the projectors to light up cyclically in the order of 1, 2, 3, 4, with each emission window lasting approximately 1 millisecond. An industrial camera with a global shutter is selected, and the shutter trigger signal is controlled by the same FPGA. Its exposure pulse width is narrowed to 0.4 milliseconds, and the center moment of each exposure is aligned with the middle of the emission pulses of the four projectors through a precision programmable delay line. With this arrangement, ideally, each time the shutter opens, only the light directly projected onto the screen by one projector can be received. However, the actual device is not an ideal switch. After the high-power LED is turned off, there is a microsecond-level phosphorescence trail, and the shutter curtain movement also has nanosecond-level jitter, causing the leading and trailing edges of the exposure window to capture the weak light flux of adjacent projectors that have not yet completely turned off or have just begun to rise, resulting in inter-frame light leakage residue.

[0024] In a preferred embodiment, in step S3, the four consecutive frames captured by the camera are considered as a spatiotemporally mixed observation sequence. For a pixel on the screen, the brightness values ​​obtained from the four shutter shots are arranged into a column vector y, where the k-th element corresponds to the k-th exposure reading. Let the pure individual illumination of the four projectors at that point be a column vector c, where the i-th component is the ideal brightness that the camera should receive when only projection i is illuminated without any crosstalk. In reality, due to the overlap between the shutter speed and the emission pulse, the total brightness of each exposure is the weighted integral of the luminous flux of each projection within the exposure window. This temporal mixing can be characterized by a 4×4 matrix A, whose elements... The integral proportion representing the energy contributed by projection i during the k-th exposure is only related to the known pulse shape and delay parameters. Meanwhile, crosstalk in the spatial domain is also generated by scattering between projector lenses and stray light interference caused by the screen microstructure. S2 has already calibrated a pixel-wise crosstalk matrix W using the illumination substrate atlas, with elements... This describes the proportion of stray light transfer at that pixel by projection j relative to projection i.

[0025] The formula for the hybrid model is:

[0026]

[0027] In the formula:

[0028] This is the brightness observation column vector collected from four shutter exposures, with a length of 4;

[0029] It is a time mixing matrix with a size of 4×4;

[0030] element This represents the integral weight of the k-th shutter window with respect to the intensity of the projected i light;

[0031] It is a stray crosstalk matrix from S2, with a size of 4×4, and each pixel is different;

[0032] element This represents the ratio of stray light contribution of projection j to projection i at that pixel.

[0033] Let be the column vector of pure individual illumination to be determined, with a length of 4, representing the direct brightness of each of the four projectors under ideal conditions.

[0034] The integral definition of the elements of the time mixing matrix:

[0035]

[0036] This refers to the camera shutter exposure pulse width.

[0037] It is the time profile function of the projected i-emitting pulse, which is obtained by normalizing the measured light intensity waveform.

[0038] R represents the shutter gain curve, taken as a rectangular window or the measured response.

[0039] This is the start time of the k-th exposure.

[0040] It is the time delay of the projected i-th emission pulse relative to the k-th shutter reference, which is determined by the FPGA phase setting and can be measured in advance.

[0041] In a preferred embodiment, during the secondary rejection in step S3, for each round of four original frames acquired, the observation vector y is extracted for each pixel, and the crosstalk matrix W corresponding to that pixel is read from the weight table passed in S2. The temporal mixing matrix A has already been measured by a high-speed photoelectric probe and calculated in one go during the system calibration stage, and the condition number of A is good; its inverse matrix is ​​first fixed in the video memory. For each pixel, the pure brightness of the four projectors is directly calculated and restored by solving the pure components.

[0042] If W approaches singularity due to severe obstruction of the projection light path, direct inversion would amplify sensor noise. In this case, regularization is automatically switched to suppress solution oscillation. After traversing all pixels, four clean component images stripped of temporal light leakage and spatial crosstalk are obtained, which can be directly fed into S4 for pre-compensation curve fitting.

[0043] The formula for directly calculating the pure component is:

[0044]

[0045] in: and These are the inverse matrices of the crosstalk matrix and the time mixing matrix, respectively, and both are required to be invertible and numerically stable.

[0046] The formula for regularization at singular pixels is as follows:

[0047]

[0048] In the formula:

[0049] This is the regularization constant, usually taken to be around 0.001.

[0050] It is a 4×4 identity matrix. This formula can be found in... When the condition number is too large, a robust approximate solution is provided to avoid excessive amplification of noise.

[0051] In a preferred embodiment, in step S4, after receiving the four pure single-image component images output by S3, the current brightness combination of the point under the four projectors is extracted pixel by pixel and compared with the ideal brightness combination required by the target image channel by channel.

[0052] Measurements were performed using a 32-level grayscale bar with uniform intervals from pure black to pure white. A clean component was captured at each level to establish the measured response curve for each pixel projected onto the screen. The difference between the measured curve and the ideal gamma 2.2 straight line was calculated to deduce the compensation offset required for each input code value, forming a pre-compensation curve in the shape of an inverse function. The entire curve was encapsulated in a lookup table with 1024 entries, each with a precision of 16-bit unsigned integer. The horizontal axis represented the original grayscale to be displayed, and the vertical axis represented the compensated output grayscale. All pixel lookup tables were merged into a 2D texture of the same width as the screen, and a fixed area was permanently allocated in the GPU memory for the S5 rendering pipeline to directly sample and call using pixel shaders.

[0053] In a preferred embodiment, in step S5, before outputting each frame, the rendering engine binds the original image as a texture to the input slot of the pixel shader. When the shader executes, it first retrieves the corresponding 1024-entry lookup table from the pre-compensation texture in S4 according to the current pixel coordinates, looks up the compensation value using the grayscale value of the original pixel as the index, and then multiplies this compensation value by the ratio of the current ambient light probe reading to the calibration reference. The ambient light probe is installed in the recessed areas of the four corners of the screen frame, and transmits a set of illuminance readings via the iC bus every 200ms. The four corner readings are weighted and averaged to remove pulse glitches, resulting in a slowly changing overall bias coefficient. This coefficient acts as a uniform multiplier on the entire pre-compensation texture, causing the compensation amplitude to increase linearly with increasing ambient light and decrease linearly with decreasing ambient light. After completing the multiplication operation, the shader writes the final pixel value to the frame buffer for use in the next stage, S6, feathering and blending.

[0054] In a preferred embodiment, in step S6, the ratio of the distance from each pixel within the overlap band width region to the boundaries of the two projectors determines its mixing weight. The weight calculation uses a cosine transition curve. The weight of the left projection smoothly decreases from 1.0 at the left boundary of the overlap region to 0 at the right boundary, while the weight of the right projection changes synchronously in the opposite direction. The sum of the two weights is always 1. The transition curve truncates half a cycle of the cosine function, with a starting phase of 0 and an ending phase of π, replacing the linear gradient to eliminate the brightness ridge at the center of the overlap band. The mixed signals are aligned with the blanking interval by the frame synchronization board and synchronously output to the four projectors via the DVI interface. The final image presented on the screen is a complete image with consistent brightness and color.

[0055] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0056] 1. In this invention, by utilizing a pre-established stray light crosstalk weight table and a high-speed rotating synchronous acquisition link, the instantaneous individual images of four projectors can be completely captured in each round of four-frame shooting. Through the joint inverse operation of the temporal mixing matrix and the spatial crosstalk matrix, residual light caused by device switching trailing, shutter shake, and screen mutual reflection is removed in one go, resulting in a pure component free of stray light. This deterministically decomposes the multi-source brightness originally superimposed in each frame into independent channels, ensuring that the input for subsequent pre-compensation curve calculations is not contaminated by crosstalk, preventing crosstalk errors from being further amplified during the pre-compensation stage, increasing the upper limit of the signal-to-noise ratio of the entire correction link, and improving the overall suppression effect.

[0057] 2. In this invention, the acquisition of the pure component allows the pre-compensated texture to be baked into a lookup table texture and resident in video memory in one go. The rendering pipeline only needs to complete the lookup table and multiply-accumulate operations, without the need for online iterative solving, which significantly reduces the computational overhead of real-time processing. At the same time, the illuminance readings continuously returned by the ambient light probe are applied to the pre-compensated texture with a uniform bias coefficient, enabling the system to dynamically track changes in ambient stray light without changing the shape of the compensation curve. The compensation amplitude scales linearly with the ambient light intensity without introducing additional jitter. The cosine feathering fusion of the projection overlap area acts on the signals that have been stripped of stray light and pre-compensated, and is only responsible for bridging the geometric seams. This fundamentally improves the brightness uniformity and color consistency of the fusion band, thereby improving the overall display effect. Attached Figure Description

[0058] Figure 1 This is a schematic diagram illustrating the process principle of the present invention;

[0059] Figure 2 This is a schematic diagram of the FPGA data acquisition and processing platform structure in this invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0061] Reference Figure 1-2 A method for suppressing stray light in screen images under multi-source projection, the method includes the following steps:

[0062] S1: Turn on the projector and project a full white image. Use a calibration camera to photograph the screen to obtain the light intensity attenuation distribution of each projector at each pixel position.

[0063] S2: Based on the illumination base map obtained in step S1, perform detailed calculations on each position on the screen, calculate the energy superposition ratio when multiple light sources illuminate simultaneously, and obtain the stray light crosstalk weight table after summarizing and organizing the data.

[0064] S3: Make all projectors flash alternately at a frequency higher than the display frame rate to emit test light. The camera shutter is strictly synchronized with the light emission window of each projector to capture the instantaneous image of each projector. After capturing, call the crosstalk weight table obtained in step S2 to remove the light leakage residue between frames.

[0065] S4: Compare the pure component obtained in step S3 with the target ideal image point by point, calculate the pre-compensation curve in the form of an inverse function, and process the pre-compensation curve to make it into a compensation texture that can be directly sampled and rendered in real time.

[0066] S5: During playback, each frame of the image is pre-distorted according to the compensation texture of S4, and the readings of the ambient light probe near the screen are received at the same time.

[0067] S6: Receive the multi-channel images after pre-distortion in step S5, perform brightness feathering attenuation in the projection overlap area to make the blending band smooth and seamless, and then output to the projector array to complete the display of the entire screen image after stray light suppression. After that, the entire stray light suppression process of the screen image under multi-source projection can be ended.

[0068] In step S1, after the multi-source projection system is set up, the color temperature of all projectors is first unified to around 6500K through white balance calibration. All ambient light sources are turned off, and a standard white board is attached to the screen surface for benchmark calibration. The camera is fixed at the viewer's viewpoint. During shooting, a completely black frame is first captured as the dark current benchmark. Then, the projectors are turned on to project a completely white image. Four frames are captured separately for each projector, and the median average is taken to suppress random noise. Each white field image after dark field subtraction is directly recorded as the illumination base map corresponding to that projector. The spatial resolution of all layers is consistent with the final display resolution.

[0069] In step S2, four sets of brightness readings corresponding to each pixel position are extracted from the atlas, corresponding to the measurement values ​​when the four projectors are lit individually. Utilizing the diffuse reflection symmetry of the screen, the brightness disturbance of a single projector at a certain pixel is attributed to the local illuminance gradient formed by other projected light spots near that pixel. The contribution ratio of direct light from each projector to secondary scattering from the screen is fixed using the idea of ​​light tracing, thus obtaining a 4×4 crosstalk weight matrix. Finally, these are summarized into a stray light crosstalk weight table with the same resolution as the screen, which is directly given to S3 as the prior matrix for separation.

[0070] In step S2, the formula for calculating the elements of the crosstalk weight matrix is:

[0071]

[0072] In the formula:

[0073] Index j and i refer to the projector affected by crosstalk and the projector that is the source of stray light, respectively;

[0074] Index k is the loop variable used to iterate through all four projectors when summing the denominators. The values ​​of the three indices are all between 1 and 4.

[0075] This represents the brightness value recorded by the S1 illumination atlas at the current pixel when only the projected j pixel is lit, expressed in digit levels output by the camera. The same meaning and corresponding to the projector pointed to by the summation variable k;

[0076] The local bidirectional reflectivity of the screen at this pixel describes the light scattering efficiency from the projection j outgoing direction to the projection i receiving direction. This value is obtained by considering the local bidirectional reflectivity of the pixel in the S1 illumination substrate map. Spatial gradient and The spatial gradient is calculated jointly, reflecting the directional scattering characteristics of the screen microstructure for incident light. The denominator contains... Similarly, this applies to each projector k.

[0077] It is the correction factor for the solid angle subtended by the projection lens i for that pixel, calculated based on the principal direction of projection j. Based on the projector's optical center position, the spatial coordinates of the pixel on the screen, and the screen surface normal vector, the result is obtained through geometric relationships. Calculation method and same.

[0078] Summation term in denominator This represents the weighted sum of the scattered light flux from all projections along the direction of projection i, and is used as the normalization reference.

[0079] In step S3, when setting up the synchronous acquisition link, four DLP projectors of the same model are used, and their LED driver boards are connected to a 240Hz square wave timing generated by an FPGA.

[0080] The four enable signals output by the FPGA are sequentially offset by 90 degrees, causing the projector to light up in a cyclical manner, with each emission window lasting approximately 1 millisecond. The camera is an industrial camera with a global shutter, and the shutter trigger signal is controlled by the same FPGA.

[0081] In step S3, the four consecutive frames captured by the camera are considered as a spatiotemporally mixed observation sequence. For a certain pixel on the screen, the brightness values ​​obtained from the four shutter speeds are arranged into a column vector y, with the k-th element corresponding to the k-th exposure reading.

[0082] Let the pure individual illumination of the four projectors at this point be a column vector c. The i-th component is the ideal brightness that the camera should receive when only projector i is illuminated and there is no crosstalk. Its elements are... The integral weight representing the energy contributed by projection i during the k-th exposure is only related to the known pulse shape and delay parameters. A pixel-wise crosstalk matrix W is calibrated using the illumination substrate atlas, with elements... This describes the proportion of stray light transfer at that pixel by projection j relative to projection i.

[0083] The formula for the hybrid model is:

[0084]

[0085] In the formula:

[0086] This is the brightness observation column vector collected from four shutter exposures, with a length of 4;

[0087] It is a time mixing matrix with a size of 4×4;

[0088] element This represents the integral weight of the k-th shutter window with respect to the intensity of the projected i light;

[0089] It is a stray crosstalk matrix from S2, with a size of 4×4, and each pixel is different;

[0090] element This represents the ratio of stray light contribution of projection j to projection i at that pixel.

[0091] Let be the column vector of pure individual illumination to be determined, with a length of 4, representing the direct brightness of each of the four projectors under ideal conditions.

[0092] The integral definition formula for the elements of the time mixing matrix is:

[0093]

[0094] This refers to the camera shutter exposure pulse width.

[0095] It is the time profile function of the projected i-emitting pulse, which is obtained by normalizing the measured light intensity waveform.

[0096] R represents the shutter gain curve, taken as a rectangular window or the measured response.

[0097] This is the start time of the k-th exposure.

[0098] It is the time delay of the projected i-th emission pulse relative to the k-th shutter reference, which is determined by the FPGA phase setting and can be measured in advance.

[0099] In step S3, during the secondary removal process, for each round of four original frames acquired, the observation vector y is extracted for each pixel, and the crosstalk matrix W corresponding to that pixel is read from the weight table passed in S2. The temporal mixing matrix A is calculated in one go during the system calibration phase by measuring the pulse profile using a high-speed photoelectric probe. The pure brightness of the four projectors is directly calculated and restored by solving the pure components directly for each pixel.

[0100] If W approaches singularity due to severe obstruction of the projected light path, switch to regularization to suppress oscillations in the solution. After traversing all pixels, four clean component images stripped of temporal light leakage and spatial crosstalk are obtained, which can be directly fed into the subsequent step S4 for pre-compensation curve fitting.

[0101] The formula for directly calculating the pure component is:

[0102]

[0103] in: and These are the inverse matrices of the crosstalk matrix and the time mixing matrix, respectively.

[0104] The regularization formula for singular pixels is:

[0105]

[0106] In the formula:

[0107] This is the regularization constant, usually taken to be around 0.001.

[0108] It is a 4×4 identity matrix. This formula can be found in... When the condition number is too large, a robust approximate solution is provided to avoid excessive amplification of noise.

[0109] In step S4, after receiving the four pure single-image component images output by S3, the current brightness combination of the point under the four projectors is extracted pixel by pixel and compared with the ideal brightness combination required by the target image channel by channel.

[0110] The measurement uses a grayscale bar with 32 uniformly spaced levels ranging from pure black to pure white. A pure component is captured at each level to establish the measured response curve of each device projected at that pixel. The difference between the measured curve and the ideal gamma 2.2 straight line is calculated to deduce the compensation offset required for each input code value, forming a pre-compensation curve with an inverse function shape.

[0111] In step S5, before outputting each frame, the rendering engine binds the original image as a texture to the input slot of the pixel shader. During shader execution, it first retrieves the corresponding 1024-entry lookup table from the pre-compensation texture in S4 based on the current pixel coordinates. Using the grayscale value of the original pixel as an index, it looks up the compensation value and then multiplies this compensation value by the ratio of the current ambient light probe reading to the calibration reference. The ambient light probes are installed in the recessed areas of the four corners of the screen frame. Every 200ms, they transmit a set of illuminance readings via the iC bus. The four corner readings are weighted and averaged to remove pulse glitches, resulting in a slowly changing overall bias coefficient. This coefficient acts as a uniform multiplier on the entire pre-compensation texture, causing the compensation amplitude to increase linearly with ambient light enhancement and decrease linearly with ambient light reduction. After completing the multiplication operation, the shader writes the final pixel value to the frame buffer for use in the next stage, S6, feathering and blending.

[0112] In step S6, the weight calculation uses a cosine transition curve. The weight of the left projection smoothly decreases from 1.0 at the left boundary of the overlapping area to 0 at the right boundary, while the weight of the right projection changes synchronously in the opposite direction. The sum of the two weights is always 1.

[0113] The transition curve is truncated to half a cycle of the cosine function, with the starting phase at 0 and the ending phase at π. The linear gradient is replaced to eliminate the brightness ridge at the center of the overlapping zone.

[0114] Example 1: Multilayer microprism composite structure curtain:

[0115] The screen adopts a four-layer stacked vertical structure, consisting of a gradient refractive index film, a microprism array layer, a flexible substrate layer, and a bottom light-absorbing coating, arranged sequentially from the viewer's side towards the wall.

[0116] The outermost layer is a graded refractive index film system, composed of five layers of silicon dioxide and niobium pentoxide sputtered alternately. The thickness of a single layer is on the order of hundreds of nanometers, and the total film thickness is about 1.2 micrometers. The refractive index of the film system gradually increases from 1.46 at the bottom layer to 1.78 at the top layer, forming an angle filter layer with high vertical transmittance and high oblique reflectivity. When the viewer is directly in front of the screen, light in the line of sight passes through the film system with almost no loss; stray light from adjacent cameras at an oblique angle is partially reflected back into the prism layer because the incident angle deviates from the interference design conditions of the film system.

[0117] Beneath the film system lies a microprism array layer made of UV-curable acrylic resin, formed on a flexible substrate using a roller embossing process. This layer is filled with linear prisms with a period of 0.4 mm, each extending horizontally with a non-isosceles triangular cross-section. The slope facing the main projector is gentle, with an angle of 22 degrees, and is coated with a silver-based enhanced reflective film with a reflectivity of no less than 92%. The slope away from the main projector is steep, with an angle of 67 degrees, and retains the low reflectivity of the resin itself without any coating. The main projection light enters the gentle slope at a small angle and is directed towards the audience seating area by the high-reflectivity film. Stray light from adjacent projects enters the prism at a large angle, passes through the upper surface, and strikes the inner wall of the steep slope. Due to the condition of total internal reflection, it is deflected towards the ceiling or penetrates the bottom of the prism into the light-absorbing layer.

[0118] The supporting substrate of the prism array layer is a flexible base layer made of polyimide film, 80 micrometers thick, with a coefficient of thermal expansion of less than 2 ppm. This provides a stable geometric reference surface for the microstructure above, preventing localized bulging and deformation of the screen caused by projector heat. The lower surface of the base layer is coated with a 30-micrometer thick carbon black-filled light-absorbing coating, achieving an absorption rate of 95% in the visible light band. This coating is responsible for capturing residual stray light that penetrates through the prism gaps and is deflected from the steep slope, ensuring that there is no excess light flux outside the final emission direction.

[0119] Correspondence between design principles and algorithm steps:

[0120] This screen design moves the crosstalk stripping operation in the stray light suppression process to the optical structure layer. The algorithm establishes a pixel-by-pixel stray light crosstalk weight matrix through S2, and then obtains the clean component through time-division acquisition and matrix inversion in S3. This embodiment utilizes an asymmetric microprism array and a gradient refractive index film system to directly perform directional filtering and stray light deflection of multiple projection rays in the spatial domain, effectively achieving diagonalization of the crosstalk matrix. Signals that would normally require a camera, FPGA, and real-time computation to separate are now embedded in the screen's micron-level geometry. Each projection channel is isolated from the others before the light reaches the viewer's eyes, eliminating the need for on-site calibration and acquisition in S1 to S3.

[0121] Directional decoupling mechanism of microprism array:

[0122] Each microprism has an isosceles triangle cross-section. The inclined plane facing the main projector has a 22-degree angle and is coated with a silver-based enhanced reflective film with a reflectivity of no less than 92%. This concentrates the main light beam onto the audience seating area, corresponding to the main energy contribution channel of the target projection to its own pixel in the crosstalk weighting matrix. The inclined plane facing away from the main projector has a 67-degree angle to maintain the low reflectivity of the resin. When stray light from adjacent projects enters the prism, it undergoes total internal reflection on this inclined plane and is then deflected towards the ceiling or captured by the bottom light-absorbing layer. This asymmetrical design physically deflects stray light out of the effective range before it enters the viewing cone, as if a miniature optical W inverse matrix is ​​implanted at each pixel location. The stray light contribution ratio is suppressed to below 0.01, and the main light retains a direct efficiency of close to 0.95.

[0123] The role of graded refractive index films:

[0124] Five alternating layers of silicon dioxide and niobium pentoxide are sputtered onto the upper surface of the microprism layer, with the refractive index gradually changing from 1.46 to 1.78, forming an angle filter that provides high transmission for perpendicular incident light and high reflectivity for oblique incident light. Optically, this film system plays a role in regulating the local bidirectional reflectivity Γ_ij in the algorithm, suppressing large-angle stray light through interference conditions while allowing normal light flux from the viewer's direction to pass through without attenuation. It further refines the directional selectivity of the microprism, ensuring that even if a small amount of scattered light attempts to enter the field of view, it is guided to the absorption path by the film system's reflection, effectively performing a smoothing filter correction on the crosstalk weighting table.

[0125] Factory preset and calibration-free characteristics:

[0126] During screen fabrication, a laser interferometer is used to scan the tilt angle of the prism's inclined plane point by point; any deviation exceeding 0.3 degrees results in the screen being cut off. This quality control method essentially burns the crosstalk weight table, calculated pixel by pixel in step S2, into the screen surface as a physical template. On-site installation only requires positioning the projector at the designed fixed incident angle; the screen's own microstructure automatically suppresses stray light, eliminating the need for repeated white field shooting in S1, weight table calculation in S2, and high-speed rotating image acquisition in S3. The clean image that the algorithm requires in step S4 is directly generated in the optical stage in this screen solution, requiring only routine edge feathering and blending afterwards, significantly reducing system complexity.

[0127] Example 2: Pixel-level dual-orientation light-absorbing grating screen:

[0128] The screen substrate is a 1.5 mm thick black polycarbonate sheet. Through micro-milling, an array of orthogonal grooves is formed on the surface. The intersections of the grooves form a rectangular funnel grid. The opening size of the funnel is 0.3 mm square. The center distance between adjacent funnels is precisely matched with the pixel pitch of the projected image.

[0129] Each funnel's four inner walls are electrostatically flocked, with 5-micron silicon carbide particles vertically attached to the surface at a density of 3,000 particles per square millimeter. The silicon carbide flock exhibits an absorption rate exceeding 97% in the visible light band, giving the funnel's inner walls a completely matte black optical appearance. Stray light incident obliquely from adjacent units strikes the inner wall or outer edge absorbing ribs at the funnel entrance and is directly absorbed by the flock; its energy is completely dissipated before entering the funnel and cannot reach the bottom.

[0130] The orientation of the funnel opening is not uniform, but rather designed differently point-by-point based on the spatial coordinates of the pixel on the screen. The normal vector of the funnel opening in the center of the screen is perpendicular to the screen surface and points directly to the optical center of the main projection lens. As the pixel coordinates move towards the edge of the screen, the opening normal vector gradually deflects, with the maximum deflection angle controlled within 18 degrees, always aligning with the optical center of the corresponding projector in that area. This orientation encoding ensures that the main light rays from each projector always enter along the funnel axis, passing through the aperture without obstruction.

[0131] A layer of titanium dioxide scattering microparticles, with a particle size distribution ranging from 200 to 400 nanometers, is hot-pressed into the bottom of the funnel, with an embedding depth of approximately 50 micrometers below the surface of the polycarbonate substrate. When the main light beam reaches the bottom along the funnel axis, the microparticle layer uniformly diffuses it into the viewing half-space in front of the screen, forming bright pixels with a wide viewing angle. The scattering layer itself does not receive any oblique stray light, thus keeping the diffused light flux pure.

[0132] The entire 120-inch screen is precisely assembled from 24 independently processed blocks. The funnel parameters within each block change continuously and gradually according to its position in the global coordinate system, and the opening angles of the edge blocks and the center blocks transition smoothly. The width of the seams between the blocks is controlled within 0.1 millimeters, making them completely indistinguishable to the naked eye at a viewing distance of 2 meters. The entire screen presents a uniform black appearance, with scattered bright spots only visible in the direction of the corresponding projector.

[0133] Equivalence of physical structure and algorithm crosstalk matrix:

[0134] This screen transforms the stray light crosstalk weight matrix into an independent array of micro-funnels at each pixel location. The geometric parameters of the funnels directly correspond to the solid angle correction factor Δω_ij and the local bidirectional reflectivity Γ_ij in the S2 formula. Through precision machining, a longitudinal and transverse light-absorbing grating is formed on a polycarbonate substrate. The flocked silicon carbide inner wall of the funnel achieves a visible light absorption rate of 97%. Oblique stray light is absorbed at the entrance and cannot reach the scattering layer. This is equivalent to constructing a physical W matrix at each pixel, where the stray light transfer ratio of its off-diagonal elements is suppressed to an extremely low level, while the main light transmittance of its diagonal elements remains high. This structure completes the filtering of the pure component c the instant the light reaches the screen, without any electronic shutter synchronization or matrix inversion operations.

[0135] Pixel-by-pixel differentiation design of opening orientation:

[0136] The funnel opening normal vector varies depending on the pixel's position on the screen. The central region points towards the optical center of the main projection lens, gradually deflecting towards the edges, with a maximum deflection angle of 18 degrees. This ensures that the main light rays from each projector can pass through the funnel without obstruction. This differentiated design precisely corresponds to the solid angle correction factor Δω_ij calculated in S2 based on the projector's optical center position, pixel coordinates, and screen normal vector. This maximizes the Δω factor for each projector within its designated area, while drastically reducing the Δω factor for other projectors due to geometric obstruction. The crosstalk weight matrix is ​​thus physically customized pixel-by-pixel. The entire screen naturally possesses a pre-built stray crosstalk weight table with the property of energy conservation, where the sum of rows equals 1. However, the energy of non-target channels is directionally absorbed rather than redistributed.

[0137] Bottom scattering layer and pure component display:

[0138] Titanium dioxide-doped scattering microparticles, with a particle size of 200 to 400 nanometers, are hot-pressed into the bottom of the funnel, ensuring uniform diffusion of the main light passing through the funnel into the viewing half-space and guaranteeing a wide viewing angle. This scattering layer is equivalent to directly mapping the pure individual illuminance c calculated in the algorithm to the visible image, without any luminous flux leakage from neighboring projects. Traditional solutions require inverse function pre-compensation in S4 to correct signals contaminated by crosstalk, but the base image provided by this screen is already highly pure. The pre-compensation curve mainly targets the projector's own gamma nonlinearity, greatly reducing the correction load. In the real-time rendering stage, S5 only needs to perform simple lookup table compensation and ambient light bias.

[0139] Partitioned processing and global continuum:

[0140] The 120-inch screen is divided into 24 sections, each with a photolithographically etched groove pattern, and then spliced ​​together with seams controlled to within 0.1 mm, invisible to the naked eye. The funnel parameters within each section continuously change according to its position in global coordinates, and the opening deflection angles of the edge and center sections transition smoothly, ensuring spatial continuity of crosstalk suppression performance. This manufacturing process is equivalent to batch-writing the S2 weight table, which originally required point-by-point calculation, onto the screen surface using a partitioned mold. The spliced ​​screen then forms a complete physical stray light suppression matrix. After the projector array is installed, the overlap blending area only requires standard cosine feathering because the signals entering the blending stage have already undergone optical separation at the screen surface level. The blending band will not exhibit color shifts or brightness contamination caused by cross-beams, resulting in a substantial improvement in image uniformity and color consistency.

[0141] As can be seen from the above, in this invention, by utilizing a pre-established stray light crosstalk weight table and a high-speed rotating synchronous acquisition link, the instantaneous individual images of four projectors can be completely captured in each round of four-frame shooting. Through the joint inverse operation of the temporal mixing matrix and the spatial crosstalk matrix, residual light caused by device switching trailing, shutter shake, and screen mutual reflection is removed in one go, resulting in a pure component free of stray light. This deterministically decomposes the multi-source brightness originally superimposed in each frame into independent channels, ensuring that the calculation input of the subsequent pre-compensation curve is not contaminated by crosstalk, preventing crosstalk errors from being further amplified by the pre-compensation stage, increasing the upper limit of the signal-to-noise ratio of the entire correction link, and improving the overall suppression effect.

[0142] In this invention, the acquisition of the pure component allows the pre-compensated texture to be baked into a lookup table texture and resident in video memory in one go. The rendering pipeline only needs to complete the lookup table and multiply-accumulate operations, without the need for online iterative solving, which significantly reduces the computational overhead of real-time processing. At the same time, the illuminance readings continuously returned by the ambient light probe are applied to the pre-compensated texture with a uniform bias coefficient, enabling the system to dynamically track changes in ambient stray light without changing the shape of the compensation curve. The compensation amplitude scales linearly with the ambient light intensity without introducing additional jitter. The cosine feathering blending of the projection overlap area acts on the signals that have been stripped of stray light and pre-compensated, and is only responsible for bridging the geometric seams. This fundamentally improves the brightness uniformity and color consistency of the blending band, thereby enhancing the overall display effect.

[0143] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" or any other variations thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0144] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for suppressing stray light in screen images under multi-source projection, characterized in that: The method includes the following steps: S1: Turn on the projector and project a full white screen. Use a calibration camera to photograph the screen to obtain the light intensity attenuation distribution of each projector at each pixel position. Organize these distribution data into a lighting base map atlas. S2: Based on the illumination substrate atlas obtained in step S1, calculate the energy superposition ratio of multiple light sources when they are simultaneously irradiated at each position on the screen, and summarize it into a stray light crosstalk weight table. S3: Have all projectors flash test light alternately at a frequency much higher than the display frame rate. The camera shutter is strictly synchronized with the light emission window of each projector through hard triggering, directly capturing the instantaneous single-image of each projector. After capturing, call the crosstalk weight table obtained in step S2 to perform secondary removal of inter-frame light leakage residue. S4: Compare the pure component obtained in S3 with the target ideal image point by point, calculate the pre-compensation curve in the form of an inverse function, and process the curve into a compensation texture that can be rendered and sampled in real time. S5: During playback, each frame of the image is pre-distorted according to the compensation texture of S4, and the readings of the ambient light probe near the screen are received at the same time. S6: Receives the pre-distorted multi-channel images from S5, performs brightness feathering attenuation in the projection overlap area to make the blending band smooth and seamless, and synchronously outputs them to the projector array to complete the full display after stray light suppression. Then the entire process can be ended.

2. The method for suppressing stray light in screen images under multi-source projection as described in claim 1, characterized in that: In step S1, after the multi-source projection system is set up, the color temperature of all projectors is first unified to around 6500K through white balance calibration. All ambient light sources are turned off, and a standard white board is attached to the screen surface for reference calibration. The camera is fixed at the viewer's viewpoint. When shooting, a completely black frame is first collected as the dark current reference. Then, the projector is turned on to project a completely white screen. Four frames are taken separately for each projector, and the median average is taken to suppress random noise. Each white field image after dark field subtraction is directly recorded as the illumination base map corresponding to that projector. The spatial resolution of all layers is consistent with the final display resolution.

3. The method for suppressing stray light in screen images under multi-source projection as described in claim 1, characterized in that: In step S2, to obtain the stray light crosstalk weights for each pixel, four sets of brightness readings corresponding to each pixel position are first extracted from the illumination base map transmitted from S1, which correspond to the measurement values ​​when the four projectors are lit individually. Utilizing the diffuse reflection symmetry of the screen, the brightness disturbance of a single projector at a certain pixel is attributed to the local illuminance gradient formed by other projected light spots near that pixel. The contribution ratio of direct light from each projector to secondary scattering from the screen is fixed using the idea of ​​light tracing, thereby obtaining a 4×4 crosstalk weight matrix. Finally, a stray light crosstalk weight table with the same resolution as the screen is compiled and directly handed over to S3 as a priori matrix for separation.

4. The method for suppressing stray light on a screen image under multi-source projection as described in claim 1, characterized in that: In step S2, the formula for calculating the elements of the crosstalk weight matrix is: In the formula: Index j and i refer to the projector affected by crosstalk and the projector that is the source of stray light, respectively; Index k is the loop variable used to iterate through all four projectors when summing the denominators. The values ​​of the three indices are all between 1 and 4. This represents the brightness value recorded by the S1 illumination atlas at the current pixel when only the projected j pixel is lit, expressed in digit levels output by the camera. The same meaning and corresponding to the projector pointed to by the summation variable k; The local bidirectional reflectivity of the screen at this pixel describes the light scattering efficiency from the projection j outgoing direction to the projection i receiving direction. This value is obtained by considering the local bidirectional reflectivity of the pixel in the S1 illumination substrate map. Spatial gradient and The spatial gradient is calculated jointly, reflecting the directional scattering characteristics of the screen microstructure for incident light. The denominator contains... Similarly, this applies to each projector k; It is the correction factor for the solid angle subtended by the projection lens i for that pixel, calculated based on the principal direction of projection j. Based on the projector's optical center position, the spatial coordinates of the pixel on the screen, and the screen surface normal vector, the result is obtained through geometric relationships. Calculation method and same; Summation term in denominator This represents the weighted sum of the scattered light flux from all projections to the direction of projection i, and is used as the normalization reference. In step S3, when setting up the synchronous acquisition link, four DLP projectors of the same model are used, and their LED driver boards are connected to a 240Hz square wave timing generated by an FPGA. The four enable signals output by the FPGA are shifted by 90 degrees in phase, causing the projector to light up sequentially in a loop, with each light-emitting window taking about 1 millisecond. The camera is an industrial camera with a global shutter, and the shutter trigger signal is controlled by the same FPGA.

5. The method for suppressing stray light in screen images under multi-source projection as described in claim 1, characterized in that: In step S3, when setting up the synchronous acquisition link, four DLP projectors of the same model are used, and their LED driver boards are connected to a 240Hz square wave timing generated by an FPGA. The four enable signals output by the FPGA are shifted by 90 degrees in phase, so that the projectors are lit in a cyclic manner in the order of 1, 2, 3, 4, with each light-emitting window being about 1 millisecond. An industrial camera with a global shutter is selected, and the shutter trigger signal is controlled by the same FPGA. Its exposure pulse width is narrowed to 0.4 milliseconds. The center time of each exposure is aligned with the middle of the light-emitting pulse of the four projectors by a precision programmable delay line.

6. The method for suppressing stray light in screen images under multi-source projection as described in claim 1, characterized in that: In step S3, the four consecutive frames of images captured by the camera are regarded as a spatiotemporal mixed observation sequence; for a certain pixel on the screen, the brightness values ​​obtained by the four shutter speeds are arranged into a column vector y, and the kth element corresponds to the kth exposure reading. Let the pure individual illumination of the four projectors at this point be a column vector c. The i-th component is the ideal brightness that the camera should receive when only projector i is illuminated and there is no crosstalk. Its elements are... The integral weight representing the energy contributed by projection i during the k-th exposure is only related to the known pulse shape and delay parameters. A pixel-wise crosstalk matrix W is calibrated using the illumination substrate atlas, with elements... This describes the proportion of stray light transferred by projection j to projection i at that pixel; The formula for the hybrid model is: In the formula: This is the column vector of brightness observations collected from four shutter exposures; It is a time mixing matrix; element This represents the integral weight of the k-th shutter window with respect to the intensity of the projected i light; It is a stray crosstalk matrix from S2, with a size of 4×4, and each pixel is different; element This represents the ratio of stray light contribution of projection j to projection i at that pixel; Let be the column vector of pure individual illumination to be determined, with a length of 4, which represents the direct brightness of each of the four projectors under ideal conditions. The integral definition of the elements of a time-mixing matrix: This refers to the camera shutter exposure pulse width; It is the time profile function of the projected i-emitting pulse, which is obtained by normalizing the measured light intensity waveform; R is the shutter gain curve, taken as a rectangular window or the measured response; This is the start time of the k-th exposure; It is the time delay of the projected i-th emission pulse relative to the k-th shutter reference, which is determined by the FPGA phase setting and can be measured in advance.

7. The method for suppressing stray light on a screen image under multi-source projection as described in claim 1, characterized in that: In step S3, during the secondary removal, for each round of four original frames acquired, the observation vector y is extracted for each pixel, and the crosstalk matrix W corresponding to that pixel is read from the weight table passed in S2; the temporal mixing matrix A has been measured by the high-speed photoelectric probe and solved in one go during the system calibration stage, and the condition number of A is good, and its inverse matrix is ​​first solidified in the video memory; for pixels, the pure brightness of the four projectors is directly calculated and restored by solving the pure components directly; If W approaches singularity due to severe obstruction of the projection light path, direct inversion will amplify sensor noise. In this case, the solution is automatically switched to regularization to suppress oscillation of the solution. After all pixels have been traversed, four clean component images stripped of time leakage and spatial crosstalk are obtained, which can be directly fed into S4 for pre-compensation curve fitting. The formula for directly calculating the pure component is: in: and These are the inverse matrices of the crosstalk matrix and the time mixing matrix, respectively, and both are required to be invertible and numerically stable. The formula for regularization at singular pixels is as follows: In the formula: This is the regularization constant, typically around 0.001; It is a 4×4 identity matrix; this formula can be used in... It provides a robust approximate solution when the condition number is too large.

8. The method for suppressing stray light in screen images under multi-source projection as described in claim 1, characterized in that: In step S4, after receiving the four pure single-image component images output by S3, the current brightness combination of the point under the four projectors is extracted pixel by pixel and compared with the ideal brightness combination required by the target image channel by channel. The measurement uses a grayscale bar with 32 uniformly spaced levels from all black to all white. A pure component is captured at each level to establish the measured response curve of each device projected at that pixel. The difference between the measured curve and the ideal gamma 2.2 straight line is calculated to deduce the compensation offset that needs to be applied for each input code value, forming a pre-compensation curve with an inverse function shape.

9. The method for suppressing stray light on a screen image under multi-source projection as described in claim 1, characterized in that: In step S5, before each frame is output, the rendering engine binds the original image to the input slot of the pixel shader in the form of a texture. When the shader is executed, it first retrieves the corresponding 1024 lookup table from the pre-compensation texture in S4 according to the current pixel coordinates, looks up the compensation value with the gray level value of the original pixel as the index, and then multiplies the compensation value by the ratio of the current ambient light probe reading to the calibration reference. An ambient light sensor is installed in the recessed area of ​​the four corners of the screen frame. Every 200ms, it transmits a set of illuminance readings via the iC bus. The four corner readings are weighted and averaged to remove pulse glitches, resulting in a slowly changing overall bias coefficient. This coefficient is used as a uniform multiplier on the entire pre-compensated texture, so that the compensation amplitude increases linearly with the enhancement of ambient light and decreases linearly with the weakening of ambient light. After the shader completes the multiplication operation, it writes the final pixel value into the frame buffer for use in the next stage of S6 feathering and blending.

10. The method for suppressing stray light in screen images under multi-source projection as described in claim 1, characterized in that: In step S6, the weight calculation adopts a cosine transition curve. The weight of the left projection smoothly decreases from 1.0 at the left boundary of the overlapping area to 0 at the right boundary, while the weight of the right projection changes synchronously in the opposite direction. The sum of the two weights is always 1. The transition curve is truncated to half a cycle of the cosine function, with the starting phase at 0 and the ending phase at π. The linear gradient is replaced to eliminate the brightness ridge at the center of the overlapping zone.