Real-time remote image rendering method and system based on multi-level-of-detail proxy

Abstract

The invention discloses a real-time remote image rendering method and system based on multi-level-of-detail proxy, and relates to the technical field of real-time rendering. The method comprises the following steps: acquiring multiple reference views of an image to be rendered, capturing geometric information and material information from the multiple reference views, and constructing a hierarchical agent structure; projecting agent data in the hierarchical agent structure to a target view through a forward mapping mode based on geometric transformation, and retaining most effective data through a shallowest depth pruning strategy; carrying out dual-stage missing region repair on a forward mapping result by combining bilateral filtering and thin structure region detection; and performing dynamic and editable re-coloring processing based on the geometric information and the material information in the repaired target view. According to the method, efficient and high-quality real-time rendering is realized through compact geometry and material information coding, dynamic re-coloring support and multi-level detail management.

Description

Technical Field

[0001] The present invention relates to the field of real-time rendering technology, and in particular to a real-time long-distance image rendering method and system based on a multi-level-of-detail agent. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the field of real-time rendering, efficient rendering of complex geometry at a distance remains a technical challenge. Three major inconsistencies in stereo image processing technology still exist: 1. Color inconsistency: Due to factors such as differences in camera parameters, changes in ambient lighting, and the reflective characteristics of object surfaces, there may be color differences between stereo image pairs, which will reduce the accuracy of stereo matching.

[0004] 2. View inconsistency: Stereo image pairs may experience global view displacement, deformation, or perspective differences. This inconsistency can be caused by differences in camera geometry or errors in the image acquisition process.

[0005] 3. Object inconsistency: At the local object level, a stereo image pair may have inconsistencies in the number, rendering, or location of objects. For example, some objects may be visible in one image but occluded in another.

[0006] Currently, real-time rendering primarily relies on image proxying and geometric simplification and proxying techniques. Commonly used geometric simplification and proxying methods include multi-card billboards and polyhedron proxies. Multi-card billboards simulate complex geometry using multi-plane textures but are unable to handle dynamic materials. Octahedron proxies can optimize view coverage but lack geometrically aware reconstruction capabilities.

[0007] Image imposter technology is an image processing technique that replaces original geometry with pre-rendered multi-view textures (such as color and depth). However, it currently suffers from drawbacks such as insufficient support for dynamic lighting, difficulty in material editing, high memory and bandwidth overhead, and visual artifacts. Traditional imposters only store static colors and cannot be recolored under dynamic lighting. Material properties such as roughness and metalness are baked into the texture and cannot be modified in real time. Multiple levels of detail (LoD) require storing the entire geometry buffer (G-buffer) for each level, resulting in wasted resources. View interpolation ignores geometric structure, resulting in occlusion errors and blurring (such as on sloped surfaces or high-frequency details).

[0008] Therefore, existing technologies face the dilemma of being unable to achieve both real-time performance and quality. A lightweight solution is needed to solve the three major inconsistencies in stereo image processing technology without significantly increasing computing overhead. Summary of the Invention

[0009] In response to the shortcomings of the existing technology, the purpose of the present invention is to provide a real-time long-distance image rendering method and system based on multi-level-of-detail proxies, which is suitable for image proxy (Impostor) rendering of distant objects. Through compact geometry and material information encoding, dynamic reshading support and multi-level detail management, efficient and high-quality real-time rendering is achieved.

[0010] In order to achieve the above object, the present invention is implemented through the following technical solutions: A first aspect of the present invention provides a real-time long-distance image rendering method based on a multi-level-of-detail agent, comprising the following steps: Obtain multiple reference views of the image to be rendered, capture geometric and material information from the multiple reference views, and construct a hierarchical proxy structure; Based on geometric transformation, the proxy data in the hierarchical proxy structure is projected to the target view through forward mapping, and the most effective data is retained through the shallowest depth pruning strategy; The forward mapping result is repaired by two-stage missing regions through combined bilateral filtering and thin structure region detection; Dynamic and editable reshading based on the geometry and material information in the repaired target view.

[0011] Furthermore, geometry includes information depth, normal and transparency, and material information includes UV coordinates and material ID.

[0012] Furthermore, the specific steps to retain the most effective data through the shallowest depth pruning strategy are: For multiple proxy data at the same pixel position, a shallow depth pruning strategy is executed: the depth value of each proxy in the target view is calculated, and the proxy data corresponding to the minimum depth value is selected as the most effective data to ensure that nearby objects are displayed first.

[0013] Furthermore, the specific steps of performing two-stage missing area repair on the forward mapping result by combining bilateral filtering and thin structure area detection are as follows: Perform joint bilateral filtering on the result blocks of the forward mapping; Detect thin structure regions and inpaint missing data via anisotropic diffusion.

[0014] Furthermore, the mean and variance of the depth information are calculated, and when the depth exceeds a threshold of twice the variance, it is determined to be a thin structure area.

[0015] Furthermore, during the reshading process, proxy data at different levels are dynamically switched according to the viewing distance.

[0016] Furthermore, the specific steps for dynamic and editable recoloring based on the geometric and material information in the repaired target view are as follows: Parse the material ID and retrieve the roughness and metalness parameters from the material property library; Mix multiple material data by material ratio to prevent high-frequency aliasing during rendering.

[0017] A second aspect of the present invention provides a real-time long-distance image rendering system based on a multi-level-of-detail agent, comprising: a proxy generation module configured to obtain multiple reference views of an image to be rendered, capture geometric information and material information from the multiple reference views, and construct a hierarchical proxy structure; A forward mapping module is configured to project the proxy data in the hierarchical proxy structure to the target view by forward mapping based on the geometric transformation, and retain the most effective data by the shallowest depth pruning strategy; The hole filling module is configured to perform a two-stage missing area repair on the result of the forward mapping by combining bilateral filtering and thin structure area detection; The dynamic recoloring module is configured to perform dynamic and editable recoloring based on the geometric information and material information in the repaired target view.

[0018] A third aspect of the present invention provides a medium having a program stored thereon, which, when executed by a processor, implements the steps of the real-time long-distance image rendering method based on a multi-level-of-detail agent as described in the first aspect of the present invention.

[0019] The fourth aspect of the present invention provides a device comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the real-time long-distance image rendering method based on a multi-level-of-detail agent as described in the first aspect of the present invention are implemented.

[0020] One or more of the above technical solutions have the following beneficial effects: This paper discloses a real-time, long-distance image rendering method and system based on multi-level-of-detail proxies. This method captures geometric and material information from multiple reference views, constructing a compact hierarchical proxy structure. The proxy data is projected onto the target view based on geometric transformations, avoiding interpolation artifacts. Missing regions are repaired through a combination of bilateral filtering and thin region detection. Finally, material IDs and UV coordinates are combined to enable real-time material editing and lighting calculation, achieving efficient and high-quality real-time rendering.

[0021] To address the geometric distortion caused by traditional view interpolation, this paper introduces a forward mapping technique based on projection matrix transformation. This technique maps proxy data from the reference view space to the world coordinate system through inverse matrix operations, and then achieves precise spatial transformation using the target view projection matrix. To address conflicts in projections of multiple proxy data, a shallow depth culling algorithm is designed: within the target view space, proxy data from multiple sources at the same pixel position are sorted by depth priority, retaining only the valid data closest to the viewpoint.

[0022] To address the data loss problem after forward mapping, this paper proposes a phased restoration strategy. The first phase employs an improved joint bilateral filter to comprehensively calculate pixel weights within the spatial neighborhood. In addition to conventional spatial distance weights, it also introduces a normal consistency factor and a depth gradient factor to ensure accurate restoration of geometric feature boundaries. The second phase innovatively integrates a thin region detection algorithm. By analyzing the mean and variance of depth information, it performs hole-filling filtering within the detection area and utilizes neighborhood geometric continuity for data extrapolation and filling.

[0023] To achieve real-time interaction with material properties, this paper builds a scalable material decompression pipeline. By parsing compressed material IDs, it dynamically indexes a pre-cached library of material properties (including roughness, metallicity, and subsurface scattering parameters). Furthermore, it introduces a multi-level weighted fusion mechanism: Based on the material proportions in the multi-level cache, the material shading results within the pixel coverage area are weighted and blended, eliminating material jumps when switching between levels.

[0024] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0026] Figure 1 This is a flow chart of a method for real-time long-distance image rendering based on a multi-level-of-detail proxy in Embodiment 1 of the present invention; Figure 2 This is a diagram showing the effect of executing the method in Example 1 of the present invention. DETAILED DESCRIPTION

[0027] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations; Example 1: Embodiment 1 of the present invention provides a real-time, long-distance image rendering method based on LOD proxies. This method addresses the storage redundancy issues of traditional proxy technologies by innovatively proposing a lightweight data compression strategy. Furthermore, a rendering framework with online editable and recoloring capabilities is designed based on LOD proxies. The core of this method consists of four components: LOD proxy generation and compression, forward mapping based on the LOD proxies, information filtering of the forward mapping results, and editable recoloring based on multi-layer proxies. In the LOD proxy generation and compression component, the method performs offline baking of user-input 3D model objects to generate cached textures to represent complex object information. In the forward mapping component, the geometric and material information of the object from the target perspective is obtained based on the cached textures and information from the target view. In the information filtering component, inaccurate or missing information in the forward mapping is repaired and supplemented. In the editable recoloring component, the user is allowed to edit material information at runtime, and the edited information is promptly recolored accordingly.

[0029] like Figure 1 As shown, the specific steps include: S100: Acquire multiple reference views of the object to be rendered, capture geometric information and material information of the object from the multiple reference views, and construct a hierarchical proxy structure.

[0030] In this example, geometry includes depth, normals, and transparency information, while material information includes UV coordinates and material IDs. UV coordinates define how to query 2D textures for material properties that vary with surface position and are crucial for accurate material mapping. Material IDs are used to distinguish different parts of a 3D model and associate them with different material properties, making them a crucial tool for managing multi-material models.

[0031] In one specific implementation, a high-precision 3D model (such as a triangular mesh) pre-stored on the user's disk is input. The 3D model's spatial extent is calculated based on the vertex distribution of the 3D model, and an object bounding sphere is constructed based on its center and radius. Six to 18 orthogonal projection reference views are evenly distributed around the object bounding sphere to construct a basic image proxy (imploster). This proxy stores geometric information such as depth, normals, and transparency, as well as material information such as UV coordinates and material IDs.

[0032] Offline baking process: Based on the depth information in the geometry, multiple levels of detail (LoD) are constructed for the geometry and material information in the imposter, resulting in a hierarchical proxy structure. Specifically, the input 3D model is orthogonally projected from multiple perspectives to generate a proxy texture set containing both geometry (G-buffer) and material (M-buffer). After acquiring the texture cache data from the imposter, the LOD hierarchy is constructed based on the minimum depth information between adjacent pixels in each texture.

[0033] For geometry and material information baking, we use 6-18 orthogonal cameras spaced at equal angles, with the center of the target object's bounding sphere as the origin. For each view, we render the following data into a 32-bit floating-point texture: The geometry texture (G-buffer) stores the RGBA channels of the normal (2D octahedron encoding), depth (Z value), and transparency (alpha). The material texture (M-buffer) stores the UV coordinates (16-bit floating-point compression) and the material ID (8-bit index compression) in the RG channel.

[0034] After completing offline baking of the maximum resolution geometry and material information, we then build the LoD hierarchy for the G-Buffer and M-Buffer using the following strategy: Depth: Take the minimum depth value of the 4×4 neighborhood. The minimum depth value of the adjacent pixels is taken and stored in the high-level LoD texture to maintain continuity. All subsequent geometric and material properties will be filtered based on the pixel value of the minimum depth. If the depth of a pixel in the adjacent pixels is significantly greater than the minimum depth, its properties will be discarded. Normal: Filter based on the von Mises-Fisher distribution to perform probabilistic filtering on the normal vectors to preserve the main direction. For all normals that pass the depth filter, , converting it to a certain direction on the von Mises-Fisher spherical distribution .in is a value related to roughness, Normal information for each pixel.

[0035] Material ID and UV: For all adjacent pixel information, the minimum adjacent material ID and UV coordinates are selected and encoded into a single-channel 32-bit float. Specifically, the material IDs of four adjacent pixels are compressed into a single-channel 32-bit floating-point number; the UV coordinates are normalized and mapped to the 0-1 range and encoded as 16-bit fixed-point numbers.

[0036] The above information will be queried in the editable reshading and participate in the shading calculation. At the same time, the baked cache texture and its LoD information will be compressed 2x using Block Compression (BC5 / BC7) to reduce the bandwidth pressure caused by the material and collection proxy texture set.

[0037] The real-time rendering pipeline then proceeds. Proxy data is dynamically loaded based on the viewpoint, and the final image is generated through forward mapping, hole filling, and material reshading. It supports GPU acceleration on PC, consoles, and mobile devices.

[0038] S200: Projecting the proxy data in the hierarchical proxy structure to the target view through forward mapping based on geometric transformation to avoid interpolation artifacts and retaining the most effective data through the shallowest depth pruning strategy.

[0039] S201: Projecting the proxy data in the hierarchical proxy structure to the target view through forward mapping based on the geometric transformation.

[0040] In a specific embodiment, the forward mapping processing part uses the linear transformation relationship between the reference image and the target image in the imposter to map the information in the reference image to the target perspective through forward mapping: .

[0041] in, and To pre-cache the geometry and material information in the image proxy, and Represents the geometric information and material information after projection, is the projection matrix of the imposter's reference view, is the projection matrix of the target view.

[0042] S202: Retain the most effective data through the shallowest depth pruning strategy.

[0043] In a specific embodiment, a shallow depth pruning strategy is performed on multiple proxy data at the same pixel position: the depth value of each proxy in the target view is calculated. , select the minimum depth value The corresponding proxy data is used as the most effective data to ensure that nearby objects are displayed first.

[0044] S300: Performing two-stage information screening and hole filling filtering on the results of the forward mapping. Specifically, two-stage missing area repair is performed by combining bilateral filtering and thin structure area detection to complete the filtering and completion of effective information.

[0045] S301: Perform joint bilateral filtering on the forward mapped result blocks. Perform joint bilateral filtering on the blocks, with weights including spatial distance, normal and depth consistency.

[0046] Specifically, the hole filling filter divides the target view into 8×8 pixel blocks. For each pixel in the block, the back-facing information where the angle between the normal and the view direction is greater than 90° is removed, and a joint bilateral filter is applied to smooth the geometry and material data: .

[0047] in, and Represents the smoothed geometric and material information. is the pixel that needs bilateral joint filtering, for The adjacent pixels of the block, The weight of each pixel in the block when filtering. pass and The world space distance and screen space distance (normalized distance) between the two pixels, the depth between the two pixels (the difference in depth) and the difference in normals (the dot product between the normals) are calculated.

[0048] S302: Detect thin structure areas (depth gradient > threshold) and repair missing data through anisotropic diffusion.

[0049] In this embodiment, the average value and variance of the depth information are calculated, and when the depth exceeds a threshold of twice the variance, it is determined to be a thin structure area.

[0050] Specifically, after the hole filling filter, thin object detection is performed. Using depth information, the AMD FidelityFX™ Super Resolution 2.0 Thin Object Detector identifies uncovered areas. A context-aware inpainting algorithm is then used to fill the holes, prioritizing edge structure and material continuity. In the forward mapping results, areas without any mapping information are considered uncovered.

[0051] The specific steps of the algorithm for filling holes are as follows: The bilaterally filtered texture is divided into 3x3 blocks with a step size of 1. The center point of each block is checked to see if it is a singular value compared to the rest of the points. If it is a singular value, it is filled with the average value of the rest of the pixels in the block. Otherwise, the original value is retained.

[0052] S400: Dynamic and editable reshading based on the geometry and material information in the repaired target view.

[0053] During the reshading process of this embodiment, proxy data at different levels are dynamically switched according to the viewing distance.

[0054] S401: Parse the material ID and retrieve the roughness and metalness parameters in the material attribute library.

[0055] In a specific embodiment, the reshading part can be edited, and attributes can be dynamically obtained from the pre-cached material library (Albedo, Roughness, Metallic) according to the material ID; runtime modification of parameters (such as adjusting metalness and roughness) can be supported, and the shading results can be updated in real time.

[0056] S402: Mix multi-material data according to material proportion to prevent high-frequency aliasing during rendering.

[0057] In a specific embodiment, direct lighting is calculated based on the Phong lighting model or the PBR pipeline, combining the target view normal and material properties.

[0058] Taking the calculation process of the Phong lighting model as an example, the total shading result is Diffuse+Specular. The calculation of the Diffuse part usually uses the Albedo of the material. LightColor dot(normal, lightdir), the calculation of the Specular part is usually pow(max(dot(viewDirection, reflectionDirection),0.0), 32) specularColor LightColor. In the LoD level management section, according to the view distance With the object radius Calculate the current LOD level: .

[0059] After calculating the current LOD level, the geometry and material information of the corresponding LoD level are used for shading calculations, thereby reducing the aliasing problem caused by high-frequency information flickering and improving rendering efficiency.

[0060] This embodiment supports the following application scenarios: long-distance vegetation and building rendering in open world games; real-time interaction of dynamic material surfaces in virtual reality. Figure 2As shown, compared with traditional proxy methods, the mean square error (MSE) is reduced by 4-10 times, and the LPIPS index is improved by 85%; on the RTX 4090 GPU, 18-view proxy rendering takes ≤0.55ms, and the full pipeline frame rate is >900 FPS; it supports dynamic adjustment of parameters such as metalness and roughness, and the response time is <0.5ms.

[0061] Compared to traditional image proxy technology, this invention addresses issues such as insufficient dynamic lighting support, the inability to edit materials in real time, and sudden changes in multi-level detail switching. It also significantly reduces storage overhead and improves the rendering quality and efficiency of complex objects at long distances, making it suitable for real-time interactive scenarios such as game engines and virtual reality. Compared to traditional image proxy technology, this invention achieves improvements in storage efficiency, geometric expression capabilities, and rendering capabilities. Storage efficiency is significantly improved compared to previous imposter methods, while ensuring that storage overhead does not increase excessively, while significantly improving the ability to express complex geometry. Ultimately, using image proxies, ultra-low-cost real-time rendering (>1000 fps) is achieved.

[0062] Example 2: A second embodiment of the present invention provides a real-time long-distance image rendering system based on a multi-level-of-detail proxy, comprising: a proxy generation module configured to obtain multiple reference views of an image to be rendered, capture geometric information and material information from the multiple reference views, and construct a hierarchical proxy structure; A forward mapping module is configured to project the proxy data in the hierarchical proxy structure to the target view by forward mapping based on the geometric transformation to avoid interpolation artifacts and retain the most effective data through the shallowest depth pruning strategy; The hole filling module is configured to perform a two-stage missing area repair on the result of the forward mapping by combining bilateral filtering and thin structure area detection; The dynamic recoloring module is configured to perform dynamic and editable recoloring based on the geometric information and material information in the repaired target view.

[0063] Example 3: A third embodiment of the present invention provides a medium having a program stored thereon. When the program is executed by a processor, the steps of the real-time long-distance image rendering method based on a multi-level-of-detail agent as described in the first embodiment of the present invention are implemented.

[0064] Example 4: Embodiment 4 of the present invention provides a device, including a memory, a processor, and a program stored in the memory and runnable on the processor. When the processor executes the program, the steps in the real-time long-distance image rendering method based on a multi-level-of-detail agent as described in Embodiment 1 of the present invention are implemented.

[0065] The steps involved in the above embodiments 2, 3 and 4 correspond to those in the method embodiment 1. For the specific implementation methods, please refer to the relevant description part of the embodiment 1.

[0066] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0067] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A real-time long-distance image rendering method based on a multi-level-of-detail agent, characterized in that: The following steps are involved: Obtain multiple reference views of the image to be rendered, capture geometric and material information from the multiple reference views, and construct a hierarchical proxy structure; Based on geometric transformation, the proxy data in the hierarchical proxy structure is projected to the target view through forward mapping, and the most effective data is retained through the shallowest depth pruning strategy; The forward mapping result is repaired by two-stage missing regions through joint bilateral filtering and thin structure region detection; Dynamic and editable reshading based on the geometry and material information in the repaired target view.

2. The method for real-time long-distance image rendering based on a multi-level-of-detail agent according to claim 1, wherein: Geometry includes information depth, normal and transparency, material information includes uv coordinates and material ID.

3. The method for real-time long-distance image rendering based on a multi-level-of-detail agent according to claim 1, wherein: The specific steps to retain the most effective data through the shallowest depth pruning strategy are: For multiple proxy data at the same pixel position, a shallow depth pruning strategy is executed: the depth value of each proxy in the target view is calculated, and the proxy data corresponding to the minimum depth value is selected as the most effective data to ensure that nearby objects are displayed first.

4. The method for real-time long-distance image rendering based on a multi-level-of-detail proxy according to claim 1, wherein: The specific steps of performing two-stage missing area repair on the forward mapping result by combining bilateral filtering and thin structure area detection are as follows: Perform joint bilateral filtering on the result blocks of the forward mapping; Detect thin structure regions and inpaint missing data via anisotropic diffusion.

5. The method for real-time long-distance image rendering based on a multi-level-of-detail agent according to claim 4, wherein: The mean and variance of the depth information are calculated, and when the depth exceeds a threshold of twice the variance, it is determined to be a thin structure area.

6. The method for real-time long-distance image rendering based on a multi-level-of-detail proxy according to claim 1, wherein: During the reshading process, different levels of proxy data are dynamically switched based on view distance.

7. The method for real-time long-distance image rendering based on a multi-level-of-detail proxy according to claim 6, wherein: The specific steps for dynamic and editable recoloring based on the geometric and material information in the repaired target view are as follows: Parse the material ID and retrieve the roughness and metalness parameters from the material property library; Mix multiple material data by material ratio to prevent high-frequency aliasing during rendering.

8. A real-time long-distance image rendering system based on a multi-level-of-detail agent, characterized in that: include: a proxy generation module configured to obtain multiple reference views of an image to be rendered, capture geometric information and material information from the multiple reference views, and construct a hierarchical proxy structure; A forward mapping module is configured to project the proxy data in the hierarchical proxy structure to the target view by forward mapping based on the geometric transformation, and retain the most effective data by the shallowest depth pruning strategy; The hole filling module is configured to perform a two-stage missing area repair on the result of the forward mapping by combining bilateral filtering and thin structure area detection; The dynamic recoloring module is configured to perform dynamic and editable recoloring based on the geometric information and material information in the repaired target view.

9. A computer-readable storage medium, characterized in that A plurality of instructions are stored therein, and the instructions are suitable for being loaded by a processor of a terminal device and executing the real-time long-distance image rendering method based on a multi-level-of-detail agent according to any one of claims 1 to 7.

10. A terminal device, characterized in that: The method comprises a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; and the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded by the processor and executed by the real-time long-distance image rendering method based on a multi-level-of-detail agent according to any one of claims 1 to 7.

Images