For encodes via the cimbar cli, skip over frames that will fail extract

Some cimbar frames are bad and will fail extraction due to the data
aligning itself to match the anchor pattern.

This is a design bug/quirk that doesn't matter *too* much for the
on-the-fly "encode data on one screen, decode via camera" use case.
We're constantly generating new frames, after all. But for any offline,
more "QR-code" like usage -- for example printing -- having a bad image
is much more troublesome.

The fix/workaround is to check the generated image after we generate,
and make sure it extracts. If it doesn't, we skip that frame and again.

This is currently only used the cli (and various test code), due to
potential performance and infinite-loop concerns.

src/exe/cimbar/cimbar.cpp

@@ -104,6 +104,7 @@ template <typename FilenameIterable>
 int encode(const FilenameIterable& infiles, const std::string& outpath, int ecc, int color_bits, int compression_level, bool legacy_mode, bool no_fountain)
 {
 	Encoder en(ecc, cimbar::Config::symbol_bits(), color_bits);
+	en.set_encode_id(109);
 	if (legacy_mode)
 		en.set_legacy_mode();
 	for (const string& f : infiles)

src/lib/encoder/Encoder.h

@@ -3,6 +3,7 @@
 
 #include "SimpleEncoder.h"
 #include "cimb_translator/Config.h"
+#include "extractor/Scanner.h"
 #include "serialize/format.h"
 
 #include <opencv2/opencv.hpp>
@@ -61,6 +62,13 @@ inline unsigned Encoder::encode_fountain(const std::string& filename, const std:
 		if (!frame)
 			break;
 
+		// some % of generated frames (for the current 8x8 impl)
+		// will produce random patterns that falsely match as
+		// corner "anchors" and fail to extract. So:
+		// if frame fails the scan, skip it.
+		if (!Scanner::will_it_scan(*frame))
+			continue;
+
 		if (!on_frame(*frame, i))
 			break;
 		++i;

src/lib/encoder/test/CMakeLists.txt

@@ -27,6 +27,7 @@ add_test(encoder_test encoder_test)
 target_link_libraries(encoder_test
 
 	cimb_translator
+	extractor
 
 	correct_static
 	wirehair

src/lib/encoder/test/EncoderRoundTripTest.cpp

@@ -31,7 +31,7 @@ TEST_CASE( "EncoderRoundTripTest/testFountain.Pad", "[unit]" )
 	Encoder enc(30, 4, 2);
 	assertEquals( 1, enc.encode_fountain(inputFile, outPrefix) );
 
-	uint64_t hash = 0xeecc8800efce8c48;
+	uint64_t hash = 0xefcc8800efce8c48;
 	std::string path = fmt::format("{}_0.png", outPrefix);
 	cv::Mat encodedImg = cv::imread(path);
 	cv::cvtColor(encodedImg, encodedImg, cv::COLOR_BGR2RGB);

src/lib/encoder/test/EncoderTest.cpp

@@ -49,7 +49,7 @@ TEST_CASE( "EncoderTest/testFountain.4c", "[unit]" )
 	enc.set_legacy_mode();
 	assertEquals( 3, enc.encode_fountain(inputFile, outPrefix, 0) );
 
-	std::vector<uint64_t> hashes = {0xbb1cc62b662abfe5, 0xf586f6466a5b194, 0x8c2f0f40e6ecb08b};
+	std::vector<uint64_t> hashes = {0xbb1cc62b662abfe5, 0xf586f6466a5b194, 0x93a3830d042966e1};
 	for (unsigned i = 0; i < hashes.size(); ++i)
 	{
 		DYNAMIC_SECTION( "are we correct? : " << i )
@@ -93,7 +93,7 @@ TEST_CASE( "EncoderTest/testFountain.Compress", "[unit]" )
 	Encoder enc(30, 4, 2);
 	assertEquals( 1, enc.encode_fountain(inputFile, outPrefix) );
 
-	uint64_t hash = 0xb36b65402eec434e;
+	uint64_t hash = 0xa66a666543280e8e;
 	std::string path = fmt::format("{}_0.png", outPrefix);
 	cv::Mat img = cv::imread(path);
 	assertEquals( hash, image_hash::average_hash(img) );
@@ -131,12 +131,12 @@ TEST_CASE( "EncoderTest/testFountain.Size", "[unit]" )
 	std::string outPrefix = tempdir.path() / "encoder.fountain";
 
 	Encoder enc(30, 4, 2);
-	assertEquals( 1, enc.encode_fountain(inputFile, outPrefix, 16, 1.6, 1080) );
+	assertEquals( 1, enc.encode_fountain(inputFile, outPrefix, 16, 1.6, 1024) );
 
-	uint64_t hash = 0xbdc232c714226fe6;
+	uint64_t hash = 0xa66a666543280e8e;
 	std::string path = fmt::format("{}_0.png", outPrefix);
 	cv::Mat img = cv::imread(path);
-	assertEquals( 1080, img.rows );
-	assertEquals( 1080, img.cols );
+	assertEquals( 1024, img.rows );
+	assertEquals( 1024, img.cols );
 	assertEquals( hash, image_hash::average_hash(img) );
 }

src/lib/extractor/Scanner.h

@@ -2,6 +2,7 @@
 #pragma once
 
 #include "Anchor.h"
+#include "Corners.h"
 #include "Point.h"
 #include "ScanState.h"
 
@@ -10,7 +11,6 @@
 #include <iostream>
 #include <vector>
 
-class Corners;
 class Midpoints;
 
 class Scanner
@@ -33,6 +33,10 @@ class Scanner
 
 	static unsigned nextPowerOfTwoPlusOne(unsigned v); // helper
 
+	// external helper method
+	template <typename MAT>
+	static bool will_it_scan(const MAT& img);
+
 	// rest of public interface
 	std::vector<Anchor> scan();
 	std::vector<point<int>> scan_edges(const Corners& corners, Midpoints& mps) const;
@@ -98,6 +102,27 @@ inline unsigned Scanner::nextPowerOfTwoPlusOne(unsigned v)
 	return std::max(3U, v + 2);
 }
 
+template <typename MAT>
+inline bool Scanner::will_it_scan(const MAT& unpadded_img)
+{
+	Scanner scanner(unpadded_img);
+	std::vector<Anchor> points = scanner.scan();
+	if (points.size() < 4)
+		return false;
+
+	constexpr int limit = 50;
+	Corners corners(points);
+	if (corners.top_left().x() > limit or corners.top_left().y() > limit)
+		return false;
+	if (corners.top_right().x() < (unpadded_img.cols - limit) or corners.top_right().y() > limit)
+		return false;
+	if (corners.bottom_left().x() > limit or corners.bottom_left().y() < (unpadded_img.rows - limit))
+		return false;
+	if (corners.bottom_right().x() < (unpadded_img.cols - limit) or corners.bottom_right().y() < (unpadded_img.rows - limit))
+		return false;
+	return true;
+}
+
 template <typename MAT, typename MAT2>
 inline void Scanner::threshold_fast(const MAT& img, MAT2& out)
 {
@@ -141,10 +166,10 @@ inline void Scanner::preprocess_image(const MAT& img, MAT2& out, bool fast)
 
 template <typename MAT>
 inline Scanner::Scanner(const MAT& img, bool fast, bool dark, int skip)
-    : _dark(dark)
-    , _skip(skip? skip : std::min(img.rows, img.cols) / 60)
-    , _mergeCutoff(img.cols / 30)
-    , _anchorSize(30)
+	: _dark(dark)
+	, _skip(skip? skip : std::min(img.rows, img.cols) / 60)
+	, _mergeCutoff(img.cols / 30)
+	, _anchorSize(30)
 {
 	_img = preprocess_image(img, fast);
 }