diff --git a/fuzz/Cargo.lock b/fuzz/Cargo.lock
index d05c23a..fa35e51 100644
--- a/fuzz/Cargo.lock
+++ b/fuzz/Cargo.lock
@@ -1,6 +1,6 @@
 # This file is automatically @generated by Cargo.
 # It is not intended for manual editing.
-version = 3
+version = 4
 
 [[package]]
 name = "arbitrary"
@@ -26,7 +26,7 @@ dependencies = [
 
 [[package]]
 name = "image-webp"
-version = "0.1.2"
+version = "0.2.0"
 dependencies = [
  "byteorder-lite",
  "quick-error",
diff --git a/src/lossless_transform.rs b/src/lossless_transform.rs
index 3ec091b..c6b2e2d 100644
--- a/src/lossless_transform.rs
+++ b/src/lossless_transform.rs
@@ -350,11 +350,19 @@ pub(crate) fn apply_color_transform(
     let width = usize::from(width);
 
     for (y, row) in image_data.chunks_exact_mut(width * 4).enumerate() {
-        for (block_x, block) in row.chunks_mut(4 << size_bits).enumerate() {
-            let block_index = (y >> size_bits) * block_xsize + block_x;
-            let red_to_blue = transform_data[block_index * 4];
-            let green_to_blue = transform_data[block_index * 4 + 1];
-            let green_to_red = transform_data[block_index * 4 + 2];
+        let row_transform_data_start = (y >> size_bits) * block_xsize * 4;
+        // the length of block_tf_data should be `block_xsize * 4`, so we could slice it with [..block_xsize * 4]
+        // but there is no point - `.zip()` runs until either of the iterators is consumed,
+        // so the extra slicing operation would be doing more work for no reason
+        let row_tf_data = &transform_data[row_transform_data_start..];
+
+        for (block, transform) in row
+            .chunks_mut(4 << size_bits)
+            .zip(row_tf_data.chunks_exact(4))
+        {
+            let red_to_blue = transform[0];
+            let green_to_blue = transform[1];
+            let green_to_red = transform[2];
 
             for pixel in block.chunks_exact_mut(4) {
                 let green = u32::from(pixel[1]);