doc: update README and ARCHITECTURE with M8 milestone details

Document completion of M8 (KPM image pyramid and feature extraction in pure Rust):
- BoxFilterPyramid8u for efficient grayscale pyramid construction
- BinomialPyramid32f with scale-space interpolation
- DoG detector and orientation assignment
- Native Rust FREAK descriptor computation

Update short-term goals to clarify M9 integration milestone.

This is the final documentation update before the v0.6.0 release.

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

Author: kalwalt

ARCHITECTURE.md

@@ -40,6 +40,8 @@ The unified core library containing all AR functionality:
         -   `freak::hough` — Hough similarity voting (4D bin discretization over translation × angle × scale) for filtering matches by transformation consistency.
         -   `freak::clustering` — K-Medoids partitioning + Binary Hierarchical Clustering (BHC) vocabulary tree for fast approximate-NN search on 96-byte FREAK descriptors. Hamming distance via 24×32-bit bit-magic. Includes a byte-identical port of C++ `vision::FastRandom` / `vision::ArrayShuffle` so the BHC tree topology matches the C++ baseline given the same seed.
         -   `freak::matcher` — `FeatureStore` (points + flat descriptor buffer) and `FeatureMatcher` with three match variants: brute force, BHC-indexed (fast path), and homography-guided (spatial filter via 3×3 inverse + `tr` radius). All three apply the C++ ratio test (default 0.7) and filter by `FeaturePoint::maxima`.
+        -   `freak::image_pyramid` — image pyramid construction via `BoxFilterPyramid8u` (box filtering for 8-bit grayscale) and `BinomialPyramid32f` (32-bit floating-point binomial pyramid with scale-space interpolation).
+        -   `freak::feature_extraction` — keypoint detection via Difference-of-Gaussians (DoG), dominant orientation assignment via circular gradient voting, and FREAK descriptor computation with native Rust implementation of the FREAK binary pattern and bit-pair comparison.
 -   **Types** (`types`): core data structures (`ARHandle`, `ARParam`, etc.).
 
 ### `crates/wasm` — `webarkitlib-wasm`

Cargo.toml

@@ -9,7 +9,7 @@ exclude = [
 resolver = "2"
 
 [workspace.package]
-version = "0.5.1"
+version = "0.6.0"
 authors = ["kalwalt <github@kalwaltart.it>"]
 edition = "2021"
 description = "A high-performance, memory-safe Rust port of WebARKitLib (ARToolKit) for native and WASM."

README.md

@@ -38,14 +38,14 @@ Add `webarkitlib-rs` to your `Cargo.toml`:
 
 ```toml
 [dependencies]
-webarkitlib-rs = "0.5"
+webarkitlib-rs = "0.6"
 ```
 
 To enable the C++ FFI backend for KPM (Natural Feature Tracking):
 
 ```toml
 [dependencies]
-webarkitlib-rs = { version = "0.5", features = ["ffi-backend"] }
+webarkitlib-rs = { version = "0.6", features = ["ffi-backend"] }
 ```
 
 > When installing from crates.io, no extra setup is required — the C++
@@ -204,7 +204,7 @@ Enable the `log-helpers` feature and call the bundled initializer once in your b
 
 ```toml
 [dependencies]
-webarkitlib-rs = { version = "0.5", features = ["log-helpers"] }
+webarkitlib-rs = { version = "0.6", features = ["log-helpers"] }
 ```
 
 ```rust
@@ -338,8 +338,15 @@ The workspace contains two crates:
   - K-Medoids clustering + Binary Hierarchical Clustering vocabulary tree, including a byte-identical `FastRandom` / `ArrayShuffle` PRNG port — 889 LOC
   - `FeatureStore` + `FeatureMatcher` with three match variants (brute, BHC-indexed, homography-guided), C++-faithful ratio test (default 0.7), and `maxima` filtering — 1167 LOC
   - **Dual-mode FFI tests run in CI on Linux / macOS / Windows** with sorted-pair equality vs the C++ baseline for brute/indexed/guided matchers (the M7 milestone validation gate). Surfaced and fixed three latent cross-platform issues: macOS `libc++` linking, GCC `<limits>` include order, and ARM64 FMA tolerance in M6-2 solvers.
+- **M8 -- KPM image pyramid & feature extraction in pure Rust** (4 steps):
+  - **Step 1**: `BoxFilterPyramid8u` — box filtering for efficient 8-bit grayscale pyramid construction
+  - **Step 2**: `interpolate.h` + `BinomialPyramid32f` — binomial pyramid and interpolation utilities for scale-space analysis
+  - **Step 3**: `DoG detector` + `OrientationAssignment` — difference-of-Gaussians keypoint detection and dominant orientation estimation
+  - **Step 4**: `FREAK descriptor` + `Keyframe` — native Rust FREAK descriptor computation and keyframe pipeline
+  - Completes the KPM feature extraction and image pyramid components in pure Rust
 
 ### 🎯 Short-term Goals (toward v1.0.0)
+- **M9 -- Integrate pure-Rust KPM pipeline**: Wire M8 image pyramid and feature extraction components with M7 matching for end-to-end native Rust NFT support.
 - **Complete KPM in idiomatic Rust**: Port the remaining KPM feature extraction and matching logic to pure Rust, removing the C++ FFI dependency, and ship a working end-to-end NFT example.
 - **Enhanced Documentation**: Expand API reference with complete module-level docs, integration walkthroughs for JS/TS, and detailed usage examples.
 - **WASM Memory Management**: Improve resource cleanup when switching engines or markers in long-running browser sessions.

crates/wasm/pkg/README.md

@@ -338,8 +338,15 @@ The workspace contains two crates:
   - K-Medoids clustering + Binary Hierarchical Clustering vocabulary tree, including a byte-identical `FastRandom` / `ArrayShuffle` PRNG port — 889 LOC
   - `FeatureStore` + `FeatureMatcher` with three match variants (brute, BHC-indexed, homography-guided), C++-faithful ratio test (default 0.7), and `maxima` filtering — 1167 LOC
   - **Dual-mode FFI tests run in CI on Linux / macOS / Windows** with sorted-pair equality vs the C++ baseline for brute/indexed/guided matchers (the M7 milestone validation gate). Surfaced and fixed three latent cross-platform issues: macOS `libc++` linking, GCC `<limits>` include order, and ARM64 FMA tolerance in M6-2 solvers.
+- **M8 -- KPM image pyramid & feature extraction in pure Rust** (4 steps):
+  - **Step 1**: `BoxFilterPyramid8u` — box filtering for efficient 8-bit grayscale pyramid construction
+  - **Step 2**: `interpolate.h` + `BinomialPyramid32f` — binomial pyramid and interpolation utilities for scale-space analysis
+  - **Step 3**: `DoG detector` + `OrientationAssignment` — difference-of-Gaussians keypoint detection and dominant orientation estimation
+  - **Step 4**: `FREAK descriptor` + `Keyframe` — native Rust FREAK descriptor computation and keyframe pipeline
+  - Completes the KPM feature extraction and image pyramid components in pure Rust
 
 ### 🎯 Short-term Goals (toward v1.0.0)
+- **M9 -- Integrate pure-Rust KPM pipeline**: Wire M8 image pyramid and feature extraction components with M7 matching for end-to-end native Rust NFT support.
 - **Complete KPM in idiomatic Rust**: Port the remaining KPM feature extraction and matching logic to pure Rust, removing the C++ FFI dependency, and ship a working end-to-end NFT example.
 - **Enhanced Documentation**: Expand API reference with complete module-level docs, integration walkthroughs for JS/TS, and detailed usage examples.
 - **WASM Memory Management**: Improve resource cleanup when switching engines or markers in long-running browser sessions.

crates/wasm/pkg/package.json

@@ -5,7 +5,7 @@
                           "kalwalt \u003cgithub@kalwaltart.it\u003e"
                       ],
     "description":  "A high-performance, memory-safe Rust port of WebARKitLib (ARToolKit) for native and WASM.",
-    "version":  "0.5.1",
+    "version":  "0.6.0",
     "license":  "LGPL-3.0-or-later",
     "repository":  {
                        "type":  "git",

package.json

@@ -1,6 +1,6 @@
 {
   "name": "webarkitlib-rs",
-  "version": "0.5.1",
+  "version": "0.6.0",
    "description": "Port of WebARKitLib (ARToolKit) to Rust and WASM",
   "private": true,
   "scripts": {