# ClusterSweep *** {% embed url="" %} ### 1. INTRODUCTION **ClusterSweep** is a specialized, lightweight post-processing utility designed for the "cleanup" phase of 2D game asset production. It addresses specific artifacts common in AI-generated sprites, procedural generation, and downscaled high-resolution art. Unlike general image editors, ClusterSweep is built on **automata logic**. It does not "paint"; it mathematically enforcing grid logic and color clustering to remove noise (orphans) and unwanted gradients (blur). #### 1.1 Core Capabilities * **Orphan Destruction:** Cellular automata algorithms to identify and merge isolated pixels. * **Euclidean Color Snapping:** Hard-quantization of pixel colors to crush anti-aliasing artifacts. * **Zero-Latency Preview:** Direct buffer manipulation for instant feedback. * **Pipeline Agnostic:** Native Windows Drag-and-Drop support for PNG/JPG/BMP. *** ### 2. SYSTEM REQUIREMENTS & INSTALLATION #### 2.1 Hardware Specifications ClusterSweep is extremely low-profile. * **OS:** Windows 10 (1903+) or Windows 11. * **RAM:** 100MB available (process footprint is typically <40MB). * **GPU:** Not required (CPU-based arithmetic). * **Disk Space:** 250KB (approx). #### 2.2 Software Dependencies * **Microsoft .NET Framework 4.8:** Pre-installed on all modern Windows installations. No additional runtime downloads are required. #### 2.3 Installation / Portability ClusterSweep is **portable software**. It does not modify the Windows Registry or allow network connections. 1. Extract the ClusterSweep\_v1.1.0.zip file to any location. 2. Run ClusterSweep.exe. > **Note on Windows SmartScreen:** Because ClusterSweep is an independent tool without a globally certified corporate signature, Windows may flag it as "Unrecognized." This is normal. Click **More Info -> Run Anyway** to bypass. *** ### 3. INTERFACE OVERVIEW The GUI is divided into two primary zones: the **Control Decks (Left)** and the **Viewport (Right)**. #### 3.1 Sidebar Controls * **Status Readout:** Displays image resolution and current operation status. * **Clean Pass (Quick):** Runs a single iteration of the orphan removal algorithm. Best for preserving fine details. * **Deep Scrub:** Runs 3 recursive iterations. Best for removing heavy dithering noise. * **Sensitivity Slider (0-100%):** Controls the Euclidean distance threshold for color merging. * Low values: Only merges nearly identical colors. * High values: Aggressively reduces color count. * **Palette Snap:** Executes the gradient reduction based on the Sensitivity slider. * **Compare Toggle:** Hold-to-preview mechanism. Switches the viewport buffer to the original file while depressed. * **Reset:** Reverts the working buffer to the original file state. * **Export:** Writes the current buffer to disk as a .PNG. #### 3.2 Viewport * **Render Mode:** NearestNeighbor. The software strictly prevents bilinear filtering to ensure you see exact pixels. * **Background:** Checkerboard pattern (programmatically generated) to visualize transparency (Alpha). *** ### 4. ALGORITHMIC DOCUMENTATION Understanding the underlying logic allows for better usage of the tools. #### 4.1 Orphan Removal (The "Despeckle" Algorithm) The "Clean Pass" uses a 4-neighbor Cellular Automata rule set. * **Logic:** For every pixel P(x,y), the algorithm checks neighbors Top, Bottom, Left, Right. * **Cluster Rule:** If P differs from all adjacent neighbors (a 1x1 island), it determines the "Dominant" neighbor (currently biased slightly to the Upper/Left scanline for waterfall consistency) and adopts that color. * **Iteration:** * Quick Pass: Scans the buffer once. * Deep Scrub: Scans the buffer, writes changes, then re-scans the new buffer result two more times. This handles "chains" of noise pixels. #### 4.2 Gradient Crushing (Palette Snapping) This tool handles "Quantization" without dithering. * **Logic:** It utilizes a **Greedy Euclidean Distance** comparison. * **Process:** 1. The algorithm scans pixels linearly. 2. It maintains a dynamic cache of "Seen Colors." 3. For every new pixel, it calculates the 3D distance in RGB space: R1−R2)2 + (G1−G2)2 + (B1−B2)2. * If the distance to a "Seen Color" is less than the **Sensitivity Threshold**, the pixel is forced to match the "Seen Color." * If outside the threshold, the pixel is added to the "Seen Color" cache. *** ### 5. WORKFLOW RECOMMENDATIONS #### Scenario A: Cleaning AI Generation AI image generators (Midjourney/DALL-E) often add "noise" to pixel art prompts. 1. **Drop** the raw AI image into ClusterSweep. 2. **Slide Sensitivity** to \~15-20%. 3. Click **Palette Snap**. This flattens the subtle gradients AI adds to flat surfaces. 4. Click **Deep Scrub**. This removes the "confetti" pixels often found in empty space. 5. **Export.** #### Scenario B: Downscaling HD Art When shrinking a 1080p drawing to 64x64, resampling creates blur. 1. **Drop** the downscaled image. 2. **Slide Sensitivity** to \~30% (Aggressive). 3. Click **Palette Snap**. 4. Use **Quick Pass** manually (click 1-2 times) to tidy up edges without destroying the shape. *** ### 6. TECHNICAL ARCHITECTURE ClusterSweep achieves high performance via **Unmanaged Memory Access** (unsafe context). * **Bitmap Handling:** It bypasses the standard .NET GetPixel()/SetPixel() methods, which are slow due to overhead. * **Direct Buffer Access:** The application locks the WriteableBitmap back-buffer and manipulates the raw byte array (byte\*) directly using pointer arithmetic. * **Format:** The engine forces BGRA32 pixel format for consistent stride calculation (Width × 4 bytes) across all input file types. *** ### 7. LICENSE & LEGAL **ClusterSweep** is proprietary donationware. * **Usage:** You are free to use assets generated/cleaned by this tool in commercial or non-commercial projects (Games, Art, UI) without attribution. * **Distribution:** You may not redistribute, repackage, or sell the ClusterSweep.exe binary itself. * **Warranty:** The software is provided "as is," without warranty of any kind. **(c) 2025** [**Andras Gregori**](https://linktr.ee/agregori) **@** [**GregOrigin LLC**](https://www.gregorigin.com)**.**