Fix pixels being rounded to int values in some cases (#2228)

dlib/matrix/matrix_utilities.h

@@ -3049,7 +3049,7 @@ namespace dlib
             const M& m
         )
         {
-            pixel = static_cast<unsigned char>(m(0));
+            pixel = static_cast<P>(m(0));
         }
     };
 

dlib/test/image.cpp

@@ -1754,6 +1754,23 @@ namespace
                 << " error: " << max(abs(chip-255)) );
         }
 
+        // So the same as above, but for an image with float values that are all the same to make
+        // sure noting funny happens for float images.
+        {
+            print_spinner();
+            const long nr = 53;
+            const long nc = 67;
+            const long size = 8*9;
+            const double angle = 30*pi/180;
+
+            matrix<float> img(501,501), chip;
+            img = 1234.5;
+            chip_details details(centered_rect(center(get_rect(img)),nr,nc), size, angle);
+            extract_image_chip(img, details, chip);
+            DLIB_TEST_MSG(max(abs(chip-1234.5))==0,"nr: " << nr << "  nc: "<< nc << "  size: " << size << "  angle: " << angle 
+                << " error: " << max(abs(chip-255)) );
+        }
+
 
         {
             // Make sure that the interpolation in extract_image_chip() keeps stuff in the

dlib/test/pixel.cpp

@@ -449,7 +449,7 @@ namespace
         p_gray = 8;
         p_schar = 8;
         p_int = 8;
-        p_float = 8;
+        p_float = 8.5;
 
         p_hsi.h = 9;
         p_hsi.s = 10;
@@ -501,7 +501,7 @@ namespace
         DLIB_TEST(p_gray == 8);
         DLIB_TEST(p_schar == 8);
         DLIB_TEST(p_int == 8);
-        DLIB_TEST(p_float == 8);
+        DLIB_TEST(p_float == 8.5);
 
         DLIB_TEST(p_hsi.h == 9);
         DLIB_TEST(p_hsi.s == 10);
@@ -550,7 +550,7 @@ namespace
 
             DLIB_TEST(p_gray == 8);
             DLIB_TEST(p_int == 8);
-            DLIB_TEST(p_float == 8);
+            DLIB_TEST(p_float == 8.5);
             DLIB_TEST(p_schar == 8);
 
             DLIB_TEST(p_hsi.h == 9);
@@ -566,6 +566,7 @@ namespace
             assign_pixel(p_rgb,0);
             assign_pixel(p_lab,0);
 
+            vector_to_pixel(p_float, m_float);
             vector_to_pixel(p_gray, m_gray);
             vector_to_pixel(p_hsi, m_hsi);
             vector_to_pixel(p_rgb, m_rgb);
@@ -581,6 +582,7 @@ namespace
             DLIB_TEST(p_rgba.alpha == 7);
 
             DLIB_TEST(p_gray == 8);
+            DLIB_TEST(p_float == 8.5);
 
             DLIB_TEST(p_hsi.h == 9);
             DLIB_TEST(p_hsi.s == 10);