Automatic alignment detection system and deviation correction method for white spirit bottle cap printing patterns

Abstract

The invention discloses an automatic alignment detection system for a white spirit bottle cap printing pattern, and relates to the technical field of automatic detection. Based on a signal separation mechanism of a frequency domain component, a time domain displacement sequence is converted into frequency domain representation, and a high-frequency noise component generated by mechanical vibration of a conveyor belt is filtered through dynamic cut-off frequency; the situation that instantaneous jitter triggers error correction actions is avoided, a vibration model does not need to be pre-trained, filtering is directly achieved according to physical vibration characteristics, dependence on sensor precision is reduced, meanwhile, the invalid action frequency of an execution mechanism is reduced, and multi-scale feature decoupling processing is combined with a double-branch convolution and attention mechanism, so that the accuracy of the system is improved. Pattern structure features and interference layer features in a bottle cap image are separated from the source, in the printing process, the system calculates offset only based on the structure layer features, the problem of edge detection distortion caused by light reflection is avoided, full-automatic interference resistance is achieved, and the system is particularly suitable for the harsh conditions of high light reflection and small size of white spirit bottle caps.

Description

Technical Field

[0001] This invention relates to the field of automated inspection technology, and in particular to an automatic alignment and inspection system and correction method for printed patterns on liquor bottle caps. Background Technology

[0002] On automated production lines, automatic alignment detection and correction systems are widely used in printing, packaging and precision assembly, especially in scenarios with stringent alignment accuracy requirements, such as the printing of liquor bottle caps. Bottle caps are continuously transported to the printing station via a high-speed conveyor belt. The system needs to detect the position of the pattern on the bottle cap surface in real time and dynamically adjust the coordinates of the printing head. However, it is difficult to avoid instantaneous vibrations caused by mechanical seams, motor start-stop or load changes during the operation of the conveyor belt. This micron-level high-frequency vibration can cause the bottle cap to produce unexpected displacement at the moment of detection.

[0003] For example, CN106696462A describes a printing system and method for automatically identifying the printing position of objects. This system prints at least one initial pattern onto at least one object. The system includes an optical recognition module that acquires image data of the objects, an image processing device that converts the initial patterns into at least one printable pattern based on the image data, a printing module that prints the printable patterns onto the objects, and a transport module that carries the objects through the optical recognition module and the printing module. The method includes acquiring image data, converting the printable pattern, and printing the printable pattern. Since the initial pattern is converted into a printable pattern based on the image data of the object, the objects do not need to be pre-aligned during printing; instead, the printable pattern is printed directly, thus completely solving the printing misalignment problem caused by inaccurate object alignment.

[0004] To suppress vibration interference, current mainstream solutions employ time-domain filtering algorithms or AI-based dynamic prediction models. Some solutions, such as displacement prediction systems that combine LSTM (Long Short-Term Memory) networks, analyze historical displacement sequences to predict the position trend of the next frame, distinguishing between real displacement and instantaneous jitter. Other solutions introduce multi-sensor fusion strategies, such as installing accelerometers on conveyor belt supports and fusing vibration signals and visual data through Kalman filtering. These technologies reduce the frequency of misjudgment to some extent.

[0005] Although existing methods have improved anti-interference capabilities, the misjudgment caused by vibration has not been completely resolved. The randomness of high-frequency vibration has a coupling effect with the visual sampling frequency, and time-domain filtering may filter out the real slow offset signal. AI prediction models rely on a large amount of vibration scene training data, but the vibration modes of actual production lines are varied, and the model's generalization ability is insufficient. Multi-sensor solutions require additional hardware support, and accelerometers are prone to failure in metal dust environments, such as in liquor bottling workshops. The system may still identify instantaneous jitter as permanent offset, triggering unnecessary correction actions, leading to wear of actuators and drift of printing coordinates. Summary of the Invention

[0006] In view of the aforementioned existing problems, the present invention is proposed.

[0007] This invention provides an automatic alignment and detection system and correction method for printed patterns on liquor bottle caps, which solves the limitations of existing correction systems in scenarios such as conveyor belt vibration, curved surface reflection, and multiple patterns, including high misjudgment rate, poor generalization, and reliance on hardware calibration.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0009] In a first aspect, embodiments of the present invention provide an automatic alignment detection and correction method for printed patterns on liquor bottle caps, comprising,

[0010] Step S1: Continuously acquire surface images of the liquor bottle caps on the conveyor belt using an industrial camera;

[0011] Step S2: Perform multi-scale feature decoupling processing on a single frame image to extract the structural layer features of the pattern edges;

[0012] Step S3: Calculate offset data based on the difference between the structural layer features and the preset reference coordinates;

[0013] Step S4: Convert the offset data sequence from the time domain to the frequency domain to separate the high-frequency components and low-frequency components.

[0014] Step S5: Filter out high-frequency components and retain low-frequency components as effective offset.

[0015] Step S6: Generate correction instructions based on the effective offset and adjust the position of the printing head;

[0016] Furthermore,

[0017] Step S2, multi-scale feature decoupling processing, includes:

[0018] The high-frequency edge features of the image are extracted through the first convolutional branch to generate a structural layer feature map;

[0019] The low-frequency texture features of the image are extracted through the second convolutional branch to generate an interference layer feature map;

[0020] The feature maps of the structural layer and the interference layer are weighted and fused using the channel attention mechanism to suppress the weight of the interference layer;

[0021] Furthermore, the frequency domain transformation in step S4 employs Fourier transform;

[0022] Furthermore, during the frequency domain conversion process, the offset data sequence calculated in step S3 is converted from the time domain to the frequency domain, and the high-frequency components and low-frequency components are separated. The steps include:

[0023] Obtain the offset data sequence d[n] from step S3, where n represents the time index;

[0024] A Hanning window function preprocessing step is added to reduce spectral leakage and improve frequency domain resolution;

[0025] The formula for calculating the window function weight w[n] is:

[0026]

[0027] Where n is the time index, which is dimensionless and ranges from 0 to N-1, representing discrete time points in the sequence; N is the sequence length, which is dimensionless and determined by the sampling rate of the industrial camera; and w[n] represents the window weight coefficient, which is dimensionless and used to attenuate the edge effect of the sequence.

[0028] The window-weighted offset sequence is calculated as follows:

[0029] d w [n] = w[n]·d[n],

[0030] Where, d w [n] represents the weighted offset in μm, w[n] is the window weight coefficient, and d[n] is the original offset in μm, representing the real-time positional deviation of each bottle cap image;

[0031] Applying Fast Fourier Transform (FFT) to the time-domain sequence d w When [n] is converted to its frequency domain representation, the FFT formula is as follows:

[0032]

[0033] Where k is the frequency index, dimensionless, ranging from 0 to N-1, representing a discrete frequency point; n is the time index; N is the sequence length; and d... w [n] is the weighted offset sequence, j is the imaginary unit, dimensionless, X[k] represents the complex component in the frequency domain, and e is the natural constant;

[0034] The FFT output corresponds to the discrete frequency f. k The calculation formula is as follows:

[0035] f k =k·f s / N,

[0036] Among them, f k For discrete frequencies, the unit is Hz, k is the frequency index, and f is the frequency index. s The sampling frequency, in Hz, is determined by the frame rate of the industrial camera, with a typical value of 60Hz. N is the sequence length.

[0037] Calculate amplitude to extract spectral features:

[0038] A[k] = |X[k]|,

[0039] Where A[k] represents the amplitude in μm, and X[k] is the complex component in the frequency domain;

[0040] When separating high-frequency and low-frequency components, a judgment condition based on the vibration characteristics of the conveyor belt is added, and the cutoff frequency f is defined. c As a threshold:

[0041] f c =m·f base ,

[0042] Among them, f c The cutoff frequency is in Hz, m is the harmonic order adjustment factor used to control the frequency range, and f is the frequency. base The fundamental frequency of the conveyor belt vibration is associated with the rated speed of the drive motor, and the unit is Hz;

[0043] The separation rule is: when f k ≤f c At that time, it was classified as a low-frequency component X. low [k] = X[k], when f k >f c At that time, it was classified as a high-frequency component X. high [k] = X[k], where X low [k] represents the low-frequency component sequence, X high [k] represents the high-frequency component sequence, f k For discrete frequencies.

[0044] Furthermore, the high-frequency component filtering in step S5 is achieved by a band-stop filter, whose stopband frequency covers the harmonic components of the conveyor belt vibration fundamental frequency.

[0045] Furthermore, the preset reference coordinates are generated by matching the key points of the real-time image of the bottle cap with the key points of the pre-stored vector design draft. The key points include the geometric center of the pattern and the contour vertices.

[0046] Secondly, this invention provides an automatic alignment detection and correction system for printed patterns on liquor bottle caps, comprising,

[0047] Industrial cameras are mounted above the conveyor belt;

[0048] The image processing module is used to perform multi-scale feature decoupling processing on a single frame image, extract the structural layer features of the pattern edge, and calculate the offset data based on the difference between the structural layer features and the preset reference coordinates.

[0049] The signal processor is used to convert the offset data sequence from the time domain to the frequency domain, separate the high-frequency components from the low-frequency components, filter out the high-frequency components, and retain the low-frequency components as the effective offset.

[0050] The control module is used to generate correction commands based on the effective offset and adjust the position of the print head.

[0051] Furthermore, the image processing module includes a dual-branch convolutional neural network:

[0052] The first branch configures high-pass convolutional kernels to extract features from the structural layers;

[0053] The second branch is configured with a low-pass convolutional kernel to extract features from the interference layer.

[0054] The output is connected to the channel attention weighting unit;

[0055] Furthermore, the signal processor integrates a band-stop filter module, the stopband range of which is related to the rated speed of the conveyor belt drive motor;

[0056] The system also includes a vector graphics pre-storage module to store vector design drafts of bottle cap patterns;

[0057] The image processing module is equipped with a key point matching unit to align the real-time image with the vector design draft;

[0058] Furthermore, the vector design drafts are in SVG or PDF format, and key points are automatically located by a vector graphics parsing engine.

[0059] The beneficial effects of this invention are:

[0060] 1. This invention is based on a signal separation mechanism of frequency domain components, which converts the time domain displacement sequence into a frequency domain representation. It filters the high-frequency noise components generated by the mechanical vibration of the conveyor belt through a dynamic cutoff frequency, avoiding the triggering of incorrect correction actions by instantaneous jitter. Moreover, it does not require pre-training of a vibration model, and directly achieves filtering based on physical vibration characteristics, reducing the dependence on sensor accuracy and reducing the frequency of invalid actions of the actuator.

[0061] 2. The present invention combines multi-scale feature decoupling processing with bi-branch convolution and attention mechanism to separate pattern structure features and interference layer features in bottle cap image from the source. During the printing process, the system calculates the offset only based on the structure layer features, avoids the edge detection distortion problem caused by reflection, and achieves fully automatic anti-interference, especially suitable for the harsh conditions of high reflectivity and small size of liquor bottle caps.

[0062] 3. This invention uses key point matching of vector design drafts and pre-stores the geometric attributes of patterns as a benchmark to replace traditional manual calibration or data-driven training. When adding a bottle cap pattern, the system directly parses the vector file to generate benchmark coordinates, achieving zero-sample adaptation, avoiding production stoppages and retraining, and meeting the needs of the liquor industry for frequent updates to holiday packaging.

[0063] 4. The hardware modules and method logic of this invention are strictly corresponding. Key functions such as frequency domain filtering and vector analysis are implemented through standardized industrial components, reducing upgrade costs. The stopband range of the band-stop filter is related to the rated speed of the motor, and the vector engine supports universal formats, ensuring the parameter self-adaptation capability and compatibility of the system during long-term operation. Attached Figure Description

[0064] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is a flowchart of an automatic alignment detection and correction method for printed patterns on liquor bottle caps, as described in Example 1.

[0066] Figure 2 This is a flowchart of an automatic alignment detection and correction system for printed patterns on liquor bottle caps, as described in Example 1. Detailed Implementation

[0067] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0068] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0069] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0070] Example 1, referring to Figure 1 and Figure 2This is the first embodiment of the present invention, which provides an automatic alignment detection and correction method for printed patterns on liquor bottle caps, including the following steps:

[0071] Step S1: Continuously acquire surface images of the liquor bottle caps on the conveyor belt using an industrial camera;

[0072] Step S2: Perform multi-scale feature decoupling processing on a single frame image to extract the structural layer features of the pattern edges;

[0073] Multi-scale feature decoupling processing includes:

[0074] The high-frequency edge features of the image are extracted through the first convolutional branch to generate a structural layer feature map;

[0075] The low-frequency texture features of the image are extracted through the second convolutional branch to generate an interference layer feature map;

[0076] The feature maps of the structural layer and the interference layer are weighted and fused using the channel attention mechanism to suppress the weight of the interference layer;

[0077] Step S3: Calculate offset data based on the difference between the structural layer features and the preset reference coordinates;

[0078] The preset reference coordinates are generated by matching the key points of the real-time image of the bottle cap with the pre-stored vector design draft. The key points include the geometric center of the pattern and the contour vertices.

[0079] Step S4: Convert the offset data sequence from the time domain to the frequency domain, and separate the high-frequency components from the low-frequency components. The frequency domain conversion uses Fourier transform.

[0080] During the frequency domain transformation, the offset data sequence calculated in step S3 is transformed from the time domain to the frequency domain, and the high-frequency components and low-frequency components are separated. The steps include:

[0081] Obtain the offset data sequence d[n] from step S3, where n represents the time index;

[0082] A Hanning window function preprocessing step is added to reduce spectral leakage and improve frequency domain resolution;

[0083] The formula for calculating the window function weight w[n] is:

[0084]

[0085] Where n is the time index, which is dimensionless and ranges from 0 to N-1, representing discrete time points in the sequence; N is the sequence length, which is dimensionless and determined by the sampling rate of the industrial camera; and w[n] represents the window weight coefficient, which is dimensionless and used to attenuate the edge effect of the sequence.

[0086] The window-weighted offset sequence is calculated as follows:

[0087] d w [n] = w[n]·d[n],

[0088] Where, d w [n] represents the weighted offset in μm, w[n] is the window weight coefficient, and d[n] is the original offset in μm, representing the real-time positional deviation of each bottle cap image;

[0089] Applying Fast Fourier Transform (FFT) to the time-domain sequence d w When [n] is converted to its frequency domain representation, the FFT formula is as follows:

[0090]

[0091] Where k is the frequency index, dimensionless, ranging from 0 to N-1, representing a discrete frequency point; n is the time index; N is the sequence length; and d... w [n] is the weighted offset sequence, j is the imaginary unit, dimensionless, X[k] represents the complex component in the frequency domain, and e is the natural constant;

[0092] The FFT output corresponds to the discrete frequency f. k The calculation formula is as follows:

[0093] f k =k·f s / N,

[0094] Among them, f k For discrete frequencies, the unit is Hz, k is the frequency index, and f is the frequency index. s The sampling frequency, in Hz, is determined by the frame rate of the industrial camera, with a typical value of 60Hz. N is the sequence length.

[0095] Calculate amplitude to extract spectral features:

[0096] A[k] = |X[k]|,

[0097] Where A[k] represents the amplitude in μm, and X[k] is the complex component in the frequency domain;

[0098] When separating high-frequency and low-frequency components, a judgment condition based on the vibration characteristics of the conveyor belt is added, and the cutoff frequency f is defined. c As a threshold:

[0099] f c =m·f base ,

[0100] Among them, f c The cutoff frequency is in Hz, m is the harmonic order adjustment factor used to control the frequency range, and f is the frequency. baseThe fundamental frequency of the conveyor belt vibration is associated with the rated speed of the drive motor, and the unit is Hz;

[0101] The separation rule is: when f k ≤f c At that time, it was classified as a low-frequency component X. low [k] = X[k], when f k >f c At that time, it was classified as a high-frequency component X. high [k] = X[k], where X low [k] represents the low-frequency component sequence, X high [k] represents the high-frequency component sequence, f k For discrete frequencies.

[0102] Step S5: Filter out high-frequency components and retain low-frequency components as effective offset.

[0103] High-frequency component filtering is achieved through a band-stop filter, whose stopband frequency covers the harmonic components of the conveyor belt vibration fundamental frequency.

[0104] Step S6: Generate correction instructions based on the effective offset and adjust the position of the printing head.

[0105] The second embodiment of the present invention provides an automatic alignment detection and correction system for printed patterns on liquor bottle caps, comprising,

[0106] Industrial cameras are mounted above the conveyor belt;

[0107] The vector graphics pre-storage module stores the vector design drafts of the bottle cap pattern. The vector design drafts are in SVG or PDF format, and the key points are automatically located by the vector graphics parsing engine.

[0108] The image processing module is used to perform multi-scale feature decoupling processing on a single frame image, extract the structural layer features of the pattern edge, and calculate the offset data based on the difference between the structural layer features and the preset reference coordinates. The image processing module is equipped with a key point matching unit to align the real-time image with the vector design draft.

[0109] The image processing module includes a two-branch convolutional neural network:

[0110] The first branch configures high-pass convolutional kernels to extract features from the structural layers;

[0111] The second branch is configured with a low-pass convolutional kernel to extract features from the interference layer.

[0112] The output is connected to the channel attention weighting unit;

[0113] The signal processor is used to convert the offset data sequence from the time domain to the frequency domain, separate the high-frequency components from the low-frequency components, filter out the high-frequency components, and retain the low-frequency components as the effective offset; the signal processor integrates a band-stop filter module, whose stopband range is related to the rated speed of the conveyor belt drive motor.

[0114] The control module is used to generate correction commands based on the effective offset and adjust the position of the printing head.

[0115] It should be noted that 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for automatic alignment detection and correction of printed patterns on liquor bottle caps, characterized in that, include, Step S1: Continuously acquire surface images of the liquor bottle caps on the conveyor belt using an industrial camera; Step S2: Perform multi-scale feature decoupling processing on a single frame image to extract the structural layer features of the pattern edges; Step S3: Calculate offset data based on the difference between the structural layer features and the preset reference coordinates; Step S4: Convert the offset data sequence from the time domain to the frequency domain to separate the high-frequency components and low-frequency components. Step S5: Filter out high-frequency components and retain low-frequency components as effective offset. Step S6: Generate correction instructions based on the effective offset and adjust the position of the printing head.

2. The automatic alignment detection and correction method for printed patterns on liquor bottle caps as described in claim 1, characterized in that, The multi-scale feature decoupling process in step S2 includes: The high-frequency edge features of the image are extracted through the first convolutional branch to generate a structural layer feature map; The low-frequency texture features of the image are extracted through the second convolutional branch to generate an interference layer feature map; The feature maps of the structural layer and the interference layer are weighted and fused using the channel attention mechanism to suppress the weight of the interference layer.

3. The automatic alignment detection and correction method for printed patterns on liquor bottle caps as described in claim 1, characterized in that, The frequency domain conversion in step S4 uses Fourier transform.

4. The automatic alignment detection and correction method for printed patterns on liquor bottle caps as described in claim 3, characterized in that, During the frequency domain conversion process, the offset data sequence calculated in step S3 is converted from the time domain to the frequency domain, and the high-frequency components and low-frequency components are separated. The steps include: Obtain the offset data sequence d[n] from step S3, where n represents the time index; A Hanning window function preprocessing step is added to reduce spectral leakage and improve frequency domain resolution; The formula for calculating the window function weight w[n] is: Where n is the time index, which is dimensionless and ranges from 0 to N-1, representing discrete time points in the sequence; N is the sequence length, which is dimensionless and determined by the sampling rate of the industrial camera; and w[n] represents the window weight coefficient, which is dimensionless and used to attenuate the edge effect of the sequence. The window-weighted offset sequence is calculated as follows: d w [n]=w[n]·d[n], Where, d w [n] represents the weighted offset in μm, w[n] is the window weight coefficient, and d[n] is the original offset in μm, representing the real-time positional deviation of each bottle cap image; Applying Fast Fourier Transform (FFT) to the time-domain sequence d w When [n] is converted to its frequency domain representation, the FFT formula is as follows: Where k is the frequency index, dimensionless, ranging from 0 to N-1, representing a discrete frequency point; n is the time index; N is the sequence length; and d... w [n] is the weighted offset sequence, j is the imaginary unit, dimensionless, X[k] represents the complex component in the frequency domain, and e is the natural constant; The FFT output corresponds to the discrete frequency f. k The calculation formula is as follows: f k =k·f s / N, Among them, f k For discrete frequencies, the unit is Hz, k is the frequency index, and f is the frequency index. s The sampling frequency, in Hz, is determined by the frame rate of the industrial camera, with a typical value of 60Hz. N is the sequence length. Calculate amplitude to extract spectral features: A[k] = |X[k]|, Where A[k] represents the amplitude in μm, and X[k] is the complex component in the frequency domain; When separating high-frequency and low-frequency components, a judgment condition based on the vibration characteristics of the conveyor belt is added, and the cutoff frequency f is defined. c As a threshold: f c =m·f base , Among them, f c The cutoff frequency is in Hz, m is the harmonic order adjustment factor used to control the frequency range, and f is the frequency. base The fundamental frequency of the conveyor belt vibration is associated with the rated speed of the drive motor, and the unit is Hz; The separation rule is: when f k ≤f c At that time, it was classified as a low-frequency component X. low [k] = X[k], when f k >f c At that time, it was classified as a high-frequency component X. high [k] = X[k], where X low [k] represents the low-frequency component sequence, X high [k] represents the high-frequency component sequence, f k For discrete frequencies.

5. The automatic alignment detection and correction method for printed patterns on liquor bottle caps as described in claim 1, characterized in that, The high-frequency component filtering in step S5 is achieved by a band-stop filter, whose stopband frequency covers the harmonic components of the conveyor belt vibration fundamental frequency.

6. The automatic alignment detection and correction method for printed patterns on liquor bottle caps as described in claim 1, characterized in that, The preset reference coordinates are generated by matching the key points of the real-time image of the bottle cap with the key points of the pre-stored vector design draft. The key points include the geometric center of the pattern and the contour vertices.

7. An automatic alignment detection and correction system for printed patterns on liquor bottle caps, characterized in that, include: Industrial cameras are mounted above the conveyor belt; The image processing module is used to perform multi-scale feature decoupling processing on a single frame image and extract the structural layer features of the pattern edges; And offset data calculated based on the difference between structural layer features and preset reference coordinates; The signal processor is used to convert the offset data sequence from the time domain to the frequency domain, separate the high-frequency components from the low-frequency components, filter out the high-frequency components, and retain the low-frequency components as the effective offset. The control module is used to generate correction commands based on the effective offset and adjust the position of the printing head.

8. The automatic alignment detection and correction system for printed patterns on liquor bottle caps as described in claim 7, characterized in that, The image processing module includes a dual-branch convolutional neural network: The first branch configures high-pass convolutional kernels to extract features from the structural layers; The second branch is configured with a low-pass convolutional kernel to extract features from the interference layer. The output is connected to the channel attention weighting unit.

9. The automatic alignment detection and correction system for printed patterns on liquor bottle caps as described in claim 7, characterized in that, The signal processor integrates a band-stop filter module, whose stopband range is related to the rated speed of the conveyor belt drive motor; The system also includes a vector graphics pre-storage module to store vector design drafts of bottle cap patterns; The image processing module is equipped with a key point matching unit to align the real-time image with the vector design draft.

10. The automatic alignment detection and correction system for printed patterns on liquor bottle caps as described in claim 9, characterized in that, The vector design draft is in SVG or PDF format, and key points are automatically located by a vector graphics parsing engine.

Images