test_image_resample.py

from contextlib import contextmanager

import pytest

from PIL import Image, ImageDraw

from .helper import assert_image_equal, assert_image_similar, hopper


class TestImagingResampleVulnerability:
    # see https://github.com/python-pillow/Pillow/issues/1710
    def test_overflow(self):
        im = hopper("L")
        size_too_large = 0x100000008 // 4
        size_normal = 1000  # unimportant
        for xsize, ysize in (
            (size_too_large, size_normal),
            (size_normal, size_too_large),
        ):
            with pytest.raises(MemoryError):
                # any resampling filter will do here
                im.im.resize((xsize, ysize), Image.BILINEAR)

    def test_invalid_size(self):
        im = hopper()

        # Should not crash
        im.resize((100, 100))

        with pytest.raises(ValueError):
            im.resize((-100, 100))

        with pytest.raises(ValueError):
            im.resize((100, -100))

    def test_modify_after_resizing(self):
        im = hopper("RGB")
        # get copy with same size
        copy = im.resize(im.size)
        # some in-place operation
        copy.paste("black", (0, 0, im.width // 2, im.height // 2))
        # image should be different
        assert im.tobytes() != copy.tobytes()


class TestImagingCoreResampleAccuracy:
    def make_case(self, mode, size, color):
        """Makes a sample image with two dark and two bright squares.
        For example:
        e0 e0 1f 1f
        e0 e0 1f 1f
        1f 1f e0 e0
        1f 1f e0 e0
        """
        case = Image.new("L", size, 255 - color)
        rectangle = ImageDraw.Draw(case).rectangle
        rectangle((0, 0, size[0] // 2 - 1, size[1] // 2 - 1), color)
        rectangle((size[0] // 2, size[1] // 2, size[0], size[1]), color)

        return Image.merge(mode, [case] * len(mode))

    def make_sample(self, data, size):
        """Restores a sample image from given data string which contains
        hex-encoded pixels from the top left fourth of a sample.
        """
        data = data.replace(" ", "")
        sample = Image.new("L", size)
        s_px = sample.load()
        w, h = size[0] // 2, size[1] // 2
        for y in range(h):
            for x in range(w):
                val = int(data[(y * w + x) * 2 : (y * w + x + 1) * 2], 16)
                s_px[x, y] = val
                s_px[size[0] - x - 1, size[1] - y - 1] = val
                s_px[x, size[1] - y - 1] = 255 - val
                s_px[size[0] - x - 1, y] = 255 - val
        return sample

    def check_case(self, case, sample):
        s_px = sample.load()
        c_px = case.load()
        for y in range(case.size[1]):
            for x in range(case.size[0]):
                if c_px[x, y] != s_px[x, y]:
                    message = (
                        f"\nHave: \n{self.serialize_image(case)}\n"
                        f"\nExpected: \n{self.serialize_image(sample)}"
                    )
                    assert s_px[x, y] == c_px[x, y], message

    def serialize_image(self, image):
        s_px = image.load()
        return "\n".join(
            " ".join(f"{s_px[x, y]:02x}" for x in range(image.size[0]))
            for y in range(image.size[1])
        )

    def test_reduce_box(self):
        for mode in ["RGBX", "RGB", "La", "L"]:
            case = self.make_case(mode, (8, 8), 0xE1)
            case = case.resize((4, 4), Image.BOX)
            # fmt: off
            data = ("e1 e1"
                    "e1 e1")
            # fmt: on
            for channel in case.split():
                self.check_case(channel, self.make_sample(data, (4, 4)))

    def test_reduce_bilinear(self):
        for mode in ["RGBX", "RGB", "La", "L"]:
            case = self.make_case(mode, (8, 8), 0xE1)
            case = case.resize((4, 4), Image.BILINEAR)
            # fmt: off
            data = ("e1 c9"
                    "c9 b7")
            # fmt: on
            for channel in case.split():
                self.check_case(channel, self.make_sample(data, (4, 4)))

    def test_reduce_hamming(self):
        for mode in ["RGBX", "RGB", "La", "L"]:
            case = self.make_case(mode, (8, 8), 0xE1)
            case = case.resize((4, 4), Image.HAMMING)
            # fmt: off
            data = ("e1 da"
                    "da d3")
            # fmt: on
            for channel in case.split():
                self.check_case(channel, self.make_sample(data, (4, 4)))

    def test_reduce_bicubic(self):
        for mode in ["RGBX", "RGB", "La", "L"]:
            case = self.make_case(mode, (12, 12), 0xE1)
            case = case.resize((6, 6), Image.BICUBIC)
            # fmt: off
            data = ("e1 e3 d4"
                    "e3 e5 d6"
                    "d4 d6 c9")
            # fmt: on
            for channel in case.split():
                self.check_case(channel, self.make_sample(data, (6, 6)))

    def test_reduce_lanczos(self):
        for mode in ["RGBX", "RGB", "La", "L"]:
            case = self.make_case(mode, (16, 16), 0xE1)
            case = case.resize((8, 8), Image.LANCZOS)
            # fmt: off
            data = ("e1 e0 e4 d7"
                    "e0 df e3 d6"
                    "e4 e3 e7 da"
                    "d7 d6 d9 ce")
            # fmt: on
            for channel in case.split():
                self.check_case(channel, self.make_sample(data, (8, 8)))

    def test_enlarge_box(self):
        for mode in ["RGBX", "RGB", "La", "L"]:
            case = self.make_case(mode, (2, 2), 0xE1)
            case = case.resize((4, 4), Image.BOX)
            # fmt: off
            data = ("e1 e1"
                    "e1 e1")
            # fmt: on
            for channel in case.split():
                self.check_case(channel, self.make_sample(data, (4, 4)))

    def test_enlarge_bilinear(self):
        for mode in ["RGBX", "RGB", "La", "L"]:
            case = self.make_case(mode, (2, 2), 0xE1)
            case = case.resize((4, 4), Image.BILINEAR)
            # fmt: off
            data = ("e1 b0"
                    "b0 98")
            # fmt: on
            for channel in case.split():
                self.check_case(channel, self.make_sample(data, (4, 4)))

    def test_enlarge_hamming(self):
        for mode in ["RGBX", "RGB", "La", "L"]:
            case = self.make_case(mode, (2, 2), 0xE1)
            case = case.resize((4, 4), Image.HAMMING)
            # fmt: off
            data = ("e1 d2"
                    "d2 c5")
            # fmt: on
            for channel in case.split():
                self.check_case(channel, self.make_sample(data, (4, 4)))

    def test_enlarge_bicubic(self):
        for mode in ["RGBX", "RGB", "La", "L"]:
            case = self.make_case(mode, (4, 4), 0xE1)
            case = case.resize((8, 8), Image.BICUBIC)
            # fmt: off
            data = ("e1 e5 ee b9"
                    "e5 e9 f3 bc"
                    "ee f3 fd c1"
                    "b9 bc c1 a2")
            # fmt: on
            for channel in case.split():
                self.check_case(channel, self.make_sample(data, (8, 8)))

    def test_enlarge_lanczos(self):
        for mode in ["RGBX", "RGB", "La", "L"]:
            case = self.make_case(mode, (6, 6), 0xE1)
            case = case.resize((12, 12), Image.LANCZOS)
            data = (
                "e1 e0 db ed f5 b8"
                "e0 df da ec f3 b7"
                "db db d6 e7 ee b5"
                "ed ec e6 fb ff bf"
                "f5 f4 ee ff ff c4"
                "b8 b7 b4 bf c4 a0"
            )
            for channel in case.split():
                self.check_case(channel, self.make_sample(data, (12, 12)))

    def test_box_filter_correct_range(self):
        im = Image.new("RGB", (8, 8), "#1688ff").resize((100, 100), Image.BOX)
        ref = Image.new("RGB", (100, 100), "#1688ff")
        assert_image_equal(im, ref)


class TestCoreResampleConsistency:
    def make_case(self, mode, fill):
        im = Image.new(mode, (512, 9), fill)
        return im.resize((9, 512), Image.LANCZOS), im.load()[0, 0]

    def run_case(self, case):
        channel, color = case
        px = channel.load()
        for x in range(channel.size[0]):
            for y in range(channel.size[1]):
                if px[x, y] != color:
                    message = f"{px[x, y]} != {color} for pixel {(x, y)}"
                    assert px[x, y] == color, message

    def test_8u(self):
        im, color = self.make_case("RGB", (0, 64, 255))
        r, g, b = im.split()
        self.run_case((r, color[0]))
        self.run_case((g, color[1]))
        self.run_case((b, color[2]))
        self.run_case(self.make_case("L", 12))

    def test_32i(self):
        self.run_case(self.make_case("I", 12))
        self.run_case(self.make_case("I", 0x7FFFFFFF))
        self.run_case(self.make_case("I", -12))
        self.run_case(self.make_case("I", -1 << 31))

    def test_32f(self):
        self.run_case(self.make_case("F", 1))
        self.run_case(self.make_case("F", 3.40282306074e38))
        self.run_case(self.make_case("F", 1.175494e-38))
        self.run_case(self.make_case("F", 1.192093e-07))


class TestCoreResampleAlphaCorrect:
    def make_levels_case(self, mode):
        i = Image.new(mode, (256, 16))
        px = i.load()
        for y in range(i.size[1]):
            for x in range(i.size[0]):
                pix = [x] * len(mode)
                pix[-1] = 255 - y * 16
                px[x, y] = tuple(pix)
        return i

    def run_levels_case(self, i):
        px = i.load()
        for y in range(i.size[1]):
            used_colors = {px[x, y][0] for x in range(i.size[0])}
            assert 256 == len(used_colors), (
                "All colors should be present in resized image. "
                f"Only {len(used_colors)} on {y} line."
            )

    @pytest.mark.xfail(reason="Current implementation isn't precise enough")
    def test_levels_rgba(self):
        case = self.make_levels_case("RGBA")
        self.run_levels_case(case.resize((512, 32), Image.BOX))
        self.run_levels_case(case.resize((512, 32), Image.BILINEAR))
        self.run_levels_case(case.resize((512, 32), Image.HAMMING))
        self.run_levels_case(case.resize((512, 32), Image.BICUBIC))
        self.run_levels_case(case.resize((512, 32), Image.LANCZOS))

    @pytest.mark.xfail(reason="Current implementation isn't precise enough")
    def test_levels_la(self):
        case = self.make_levels_case("LA")
        self.run_levels_case(case.resize((512, 32), Image.BOX))
        self.run_levels_case(case.resize((512, 32), Image.BILINEAR))
        self.run_levels_case(case.resize((512, 32), Image.HAMMING))
        self.run_levels_case(case.resize((512, 32), Image.BICUBIC))
        self.run_levels_case(case.resize((512, 32), Image.LANCZOS))

    def make_dirty_case(self, mode, clean_pixel, dirty_pixel):
        i = Image.new(mode, (64, 64), dirty_pixel)
        px = i.load()
        xdiv4 = i.size[0] // 4
        ydiv4 = i.size[1] // 4
        for y in range(ydiv4 * 2):
            for x in range(xdiv4 * 2):
                px[x + xdiv4, y + ydiv4] = clean_pixel
        return i

    def run_dirty_case(self, i, clean_pixel):
        px = i.load()
        for y in range(i.size[1]):
            for x in range(i.size[0]):
                if px[x, y][-1] != 0 and px[x, y][:-1] != clean_pixel:
                    message = (
                        f"pixel at ({x}, {y}) is different:\n"
                        f"{px[x, y]}\n{clean_pixel}"
                    )
                    assert px[x, y][:3] == clean_pixel, message

    def test_dirty_pixels_rgba(self):
        case = self.make_dirty_case("RGBA", (255, 255, 0, 128), (0, 0, 255, 0))
        self.run_dirty_case(case.resize((20, 20), Image.BOX), (255, 255, 0))
        self.run_dirty_case(case.resize((20, 20), Image.BILINEAR), (255, 255, 0))
        self.run_dirty_case(case.resize((20, 20), Image.HAMMING), (255, 255, 0))
        self.run_dirty_case(case.resize((20, 20), Image.BICUBIC), (255, 255, 0))
        self.run_dirty_case(case.resize((20, 20), Image.LANCZOS), (255, 255, 0))

    def test_dirty_pixels_la(self):
        case = self.make_dirty_case("LA", (255, 128), (0, 0))
        self.run_dirty_case(case.resize((20, 20), Image.BOX), (255,))
        self.run_dirty_case(case.resize((20, 20), Image.BILINEAR), (255,))
        self.run_dirty_case(case.resize((20, 20), Image.HAMMING), (255,))
        self.run_dirty_case(case.resize((20, 20), Image.BICUBIC), (255,))
        self.run_dirty_case(case.resize((20, 20), Image.LANCZOS), (255,))


class TestCoreResamplePasses:
    @contextmanager
    def count(self, diff):
        count = Image.core.get_stats()["new_count"]
        yield
        assert Image.core.get_stats()["new_count"] - count == diff

    def test_horizontal(self):
        im = hopper("L")
        with self.count(1):
            im.resize((im.size[0] - 10, im.size[1]), Image.BILINEAR)

    def test_vertical(self):
        im = hopper("L")
        with self.count(1):
            im.resize((im.size[0], im.size[1] - 10), Image.BILINEAR)

    def test_both(self):
        im = hopper("L")
        with self.count(2):
            im.resize((im.size[0] - 10, im.size[1] - 10), Image.BILINEAR)

    def test_box_horizontal(self):
        im = hopper("L")
        box = (20, 0, im.size[0] - 20, im.size[1])
        with self.count(1):
            # the same size, but different box
            with_box = im.resize(im.size, Image.BILINEAR, box)
        with self.count(2):
            cropped = im.crop(box).resize(im.size, Image.BILINEAR)
        assert_image_similar(with_box, cropped, 0.1)

    def test_box_vertical(self):
        im = hopper("L")
        box = (0, 20, im.size[0], im.size[1] - 20)
        with self.count(1):
            # the same size, but different box
            with_box = im.resize(im.size, Image.BILINEAR, box)
        with self.count(2):
            cropped = im.crop(box).resize(im.size, Image.BILINEAR)
        assert_image_similar(with_box, cropped, 0.1)


class TestCoreResampleCoefficients:
    def test_reduce(self):
        test_color = 254

        for size in range(400000, 400010, 2):
            i = Image.new("L", (size, 1), 0)
            draw = ImageDraw.Draw(i)
            draw.rectangle((0, 0, i.size[0] // 2 - 1, 0), test_color)

            px = i.resize((5, i.size[1]), Image.BICUBIC).load()
            if px[2, 0] != test_color // 2:
                assert test_color // 2 == px[2, 0]

    def test_nonzero_coefficients(self):
        # regression test for the wrong coefficients calculation
        # due to bug https://github.com/python-pillow/Pillow/issues/2161
        im = Image.new("RGBA", (1280, 1280), (0x20, 0x40, 0x60, 0xFF))
        histogram = im.resize((256, 256), Image.BICUBIC).histogram()

        # first channel
        assert histogram[0x100 * 0 + 0x20] == 0x10000
        # second channel
        assert histogram[0x100 * 1 + 0x40] == 0x10000
        # third channel
        assert histogram[0x100 * 2 + 0x60] == 0x10000
        # fourth channel
        assert histogram[0x100 * 3 + 0xFF] == 0x10000


class TestCoreResampleBox:
    def test_wrong_arguments(self):
        im = hopper()
        for resample in (
            Image.NEAREST,
            Image.BOX,
            Image.BILINEAR,
            Image.HAMMING,
            Image.BICUBIC,
            Image.LANCZOS,
        ):
            im.resize((32, 32), resample, (0, 0, im.width, im.height))
            im.resize((32, 32), resample, (20, 20, im.width, im.height))
            im.resize((32, 32), resample, (20, 20, 20, 100))
            im.resize((32, 32), resample, (20, 20, 100, 20))

            with pytest.raises(TypeError, match="must be sequence of length 4"):
                im.resize((32, 32), resample, (im.width, im.height))

            with pytest.raises(ValueError, match="can't be negative"):
                im.resize((32, 32), resample, (-20, 20, 100, 100))
            with pytest.raises(ValueError, match="can't be negative"):
                im.resize((32, 32), resample, (20, -20, 100, 100))

            with pytest.raises(ValueError, match="can't be empty"):
                im.resize((32, 32), resample, (20.1, 20, 20, 100))
            with pytest.raises(ValueError, match="can't be empty"):
                im.resize((32, 32), resample, (20, 20.1, 100, 20))
            with pytest.raises(ValueError, match="can't be empty"):
                im.resize((32, 32), resample, (20.1, 20.1, 20, 20))

            with pytest.raises(ValueError, match="can't exceed"):
                im.resize((32, 32), resample, (0, 0, im.width + 1, im.height))
            with pytest.raises(ValueError, match="can't exceed"):
                im.resize((32, 32), resample, (0, 0, im.width, im.height + 1))

    def resize_tiled(self, im, dst_size, xtiles, ytiles):
        def split_range(size, tiles):
            scale = size / tiles
            for i in range(tiles):
                yield (int(round(scale * i)), int(round(scale * (i + 1))))

        tiled = Image.new(im.mode, dst_size)
        scale = (im.size[0] / tiled.size[0], im.size[1] / tiled.size[1])

        for y0, y1 in split_range(dst_size[1], ytiles):
            for x0, x1 in split_range(dst_size[0], xtiles):
                box = (x0 * scale[0], y0 * scale[1], x1 * scale[0], y1 * scale[1])
                tile = im.resize((x1 - x0, y1 - y0), Image.BICUBIC, box)
                tiled.paste(tile, (x0, y0))
        return tiled

    @pytest.mark.valgrind_known_error(reason="Known Failing")
    def test_tiles(self):
        with Image.open("Tests/images/flower.jpg") as im:
            assert im.size == (480, 360)
            dst_size = (251, 188)
            reference = im.resize(dst_size, Image.BICUBIC)

            for tiles in [(1, 1), (3, 3), (9, 7), (100, 100)]:
                tiled = self.resize_tiled(im, dst_size, *tiles)
                assert_image_similar(reference, tiled, 0.01)

    @pytest.mark.valgrind_known_error(reason="Known Failing")
    def test_subsample(self):
        # This test shows advantages of the subpixel resizing
        # after supersampling (e.g. during JPEG decoding).
        with Image.open("Tests/images/flower.jpg") as im:
            assert im.size == (480, 360)
            dst_size = (48, 36)
            # Reference is cropped image resized to destination
            reference = im.crop((0, 0, 473, 353)).resize(dst_size, Image.BICUBIC)
            # Image.BOX emulates supersampling (480 / 8 = 60, 360 / 8 = 45)
            supersampled = im.resize((60, 45), Image.BOX)

        with_box = supersampled.resize(dst_size, Image.BICUBIC, (0, 0, 59.125, 44.125))
        without_box = supersampled.resize(dst_size, Image.BICUBIC)

        # error with box should be much smaller than without
        assert_image_similar(reference, with_box, 6)
        with pytest.raises(AssertionError, match=r"difference 29\."):
            assert_image_similar(reference, without_box, 5)

    def test_formats(self):
        for resample in [Image.NEAREST, Image.BILINEAR]:
            for mode in ["RGB", "L", "RGBA", "LA", "I", ""]:
                im = hopper(mode)
                box = (20, 20, im.size[0] - 20, im.size[1] - 20)
                with_box = im.resize((32, 32), resample, box)
                cropped = im.crop(box).resize((32, 32), resample)
                assert_image_similar(cropped, with_box, 0.4)

    def test_passthrough(self):
        # When no resize is required
        im = hopper()

        for size, box in [
            ((40, 50), (0, 0, 40, 50)),
            ((40, 50), (0, 10, 40, 60)),
            ((40, 50), (10, 0, 50, 50)),
            ((40, 50), (10, 20, 50, 70)),
        ]:
            res = im.resize(size, Image.LANCZOS, box)
            assert res.size == size
            assert_image_equal(res, im.crop(box), f">>> {size} {box}")

    def test_no_passthrough(self):
        # When resize is required
        im = hopper()

        for size, box in [
            ((40, 50), (0.4, 0.4, 40.4, 50.4)),
            ((40, 50), (0.4, 10.4, 40.4, 60.4)),
            ((40, 50), (10.4, 0.4, 50.4, 50.4)),
            ((40, 50), (10.4, 20.4, 50.4, 70.4)),
        ]:
            res = im.resize(size, Image.LANCZOS, box)
            assert res.size == size
            with pytest.raises(AssertionError, match=r"difference \d"):
                # check that the difference at least that much
                assert_image_similar(res, im.crop(box), 20, f">>> {size} {box}")

    def test_skip_horizontal(self):
        # Can skip resize for one dimension
        im = hopper()

        for flt in [Image.NEAREST, Image.BICUBIC]:
            for size, box in [
                ((40, 50), (0, 0, 40, 90)),
                ((40, 50), (0, 20, 40, 90)),
                ((40, 50), (10, 0, 50, 90)),
                ((40, 50), (10, 20, 50, 90)),
            ]:
                res = im.resize(size, flt, box)
                assert res.size == size
                # Borders should be slightly different
                assert_image_similar(
                    res,
                    im.crop(box).resize(size, flt),
                    0.4,
                    f">>> {size} {box} {flt}",
                )

    def test_skip_vertical(self):
        # Can skip resize for one dimension
        im = hopper()

        for flt in [Image.NEAREST, Image.BICUBIC]:
            for size, box in [
                ((40, 50), (0, 0, 90, 50)),
                ((40, 50), (20, 0, 90, 50)),
                ((40, 50), (0, 10, 90, 60)),
                ((40, 50), (20, 10, 90, 60)),
            ]:
                res = im.resize(size, flt, box)
                assert res.size == size
                # Borders should be slightly different
                assert_image_similar(
                    res,
                    im.crop(box).resize(size, flt),
                    0.4,
                    f">>> {size} {box} {flt}",
                )