Features and user guide

F.I.A.T — HELP & USER GUIDE

OVERVIEW

• FIAT (Forensic Image Analysis Tool) supports professional image triage: viewing, annotating, filtering, object detection, metadata inspection, integrity checks, comparison, reporting, and more.

• Reports are saved to a “reports” folder alongside the app (you’ll be prompted to name them).

LOADING IMAGES

• Single Image… loads one file.

• Multiple Images… loads a batch.

• Drag & Drop: drop files onto the main window (single or multiple).

• Supported formats: JPG/JPEG, PNG, BMP, TIFF, WEBP (depends on your OpenCV build).

• On load, FIAT computes and stores a SHA-256 hash per file for integrity verification.

VIEWER BASICS

• Fit to View: runs automatically on image change so the whole image is visible.

• Pan: left-click drag (when no annotation tool is selected).

• Zoom: mouse wheel (smooth, anchored under cursor).

• Crop Selection: right-click drag draws a rectangle (used by Crop Image).

• Pixel Analysis: toggle on/off; shows tooltips with X, Y, RGB, and intensity and a movable magnifier.

ANNOTATIONS (MANUAL)

• Use the Annotation dropdown: Add Text, Add Circle, Undo Annotation, Clear Annotations.

• Annotations are tagged internally as “annotation” and render above the image wherever you first placed the mouse hand.

• Undo Annotation removes only the last manual annotation.

• Clear Annotations removes only manual annotations for the current image (does not clear detections).

• Annotation sizing: larger text and circle radius are supported (defaults tuned for readability).

OBJECT DETECTION (YOLOv8 ONNX)

• Choose “General (COCO-80)” then click Detect Objects.

• Detections draw red boxes and a label above the box on a soft white background.

• Uncertain labels (if any) are styled differently.

• Clear Detections (current image) removes overlays and also removes saved detection results for that image so reports will not include them.

• Thumbnails mirror detection overlays so you can track which images were processed.

• Models: We use (e.g., yolov8x.onnx) for high accuracy and performance.

FILTERS

• Available filters: Histogram Equalization, Adaptive HE (CLAHE), Sobel/Laplacian edges, High-pass, Fourier visualization, “Infrared” (colormap), Binary Threshold, Contour detection, Gabor, Brightness/Contrast, Non-Local Means denoise, RGB channel decomposition, Unsharp mask, Clear Filters.

• Applying a filter automatically clears detections and manual annotations for the current image (including saved detection data) so reports reflect the filtered state.

• Clear Filters restores the original image for that image only.

ENHANCE RESOLUTION

• Uses ESPCN x4 super-resolution if ESPCN_x4. pb, otherwise falls back to high-quality bicubic upscaling.

• The viewer preserves scale and center so the image doesn’t “jump” or zoom much unexpectedly.

SNAPSHOTS

• Take Snapshot captures the current scene (image + annotations + detections) and saves off the UI thread to avoid freezing.

• Snapshots can be saved as PNG/JPEG/BMP.

METADATA (SINGLE & MULTI)

• Single-Image Metadata shows EXIF and the stored SHA-256 line for the active image. If GPS exists, an Open in Google Maps button is shown.

• Multi-Image Metadata lists each image, metadata, SHA-256, and shows a per-image Open Map button if GPS exists.

• Generate Report creates a Word document that includes the image (processed version if present, so overlays/filters appear) and all metadata. You’ll be prompted to name the report.

VERIFY INTEGRITY

• Recomputes SHA-256 and compares against the baseline captured at load time. Results are shown via centered toast notifications.

COMPARE TWO IMAGES

• Pick Image A and B from currently loaded items; click Run Comparison.

• Metrics: SHA-256 equality, perceptual hash distances (aHash, dHash, pHash), SSIM, and Enhanced ELA (heatmap + overlay).

• Toggle ELA View switches between overlay and heatmap.

• Save Comparison Report exports a DOCX with both ELA views and all metrics; you can name the report.

CROPPING

• Right-drag to select a region; Crop Image opens a cropped viewer where you can apply grayscale, sharpening (slider), clear filters, and save a snapshot.

THUMBNAILS

• Thumbnails are clickable to switch images.

• Right click a thumbnail to delete scene and thumbnail image, open the image location or to just open the image in the scene.

• Thumbnails reflect detection overlays (and are updated when detections are cleared or re-run).

DRAG & DROP

• Drop one or more images directly into the main window to load them.

NOTIFICATIONS

• Message boxes were upgraded to centered fade-in/out toasts (self.notify). They do not block UI.

REPORT NAMING

• When generating any report, you are prompted to enter a file name. The report is saved into the “reports” folder and a toast confirms the chosen name and path.

• You do not have to rename it but it is recommended.

LIMITATIONS & TIPS

• ELA is JPEG-centric and can show hotspots from normal compression or text/edges; treat as clues, not proof.

• Perceptual hashes are robust to minor resizes/encodes, but diverge on crops, heavy filters, or local edits—use alongside SSIM and ELA.

• Larger YOLO models improve accuracy but cost performance; pick models appropriate for your hardware and timelines.

• For forensic rigor: keep originals immutable; work on copies; capture tool versions, model names, thresholds, and timestamps in reports.