diff --git a/include/hdrplus/merge.h b/include/hdrplus/merge.h
index d3aa550..bc098dc 100644
--- a/include/hdrplus/merge.h
+++ b/include/hdrplus/merge.h
@@ -5,12 +5,15 @@
 #include <cmath>
 #include "hdrplus/burst.h"
 
+#define TILE_SIZE 16
+
 namespace hdrplus
 {
 
 class merge
 {
     public:
+        int offset = TILE_SIZE / 2;
         float baseline_lambda_shot = 3.24 * pow( 10, -4 );
         float baseline_lambda_read = 4.3 * pow( 10, -6 );
 
@@ -29,7 +32,7 @@ class merge
                       std::vector<std::vector<std::vector<std::pair<int, int>>>>& alignments);
         
     private:
-        cv::Mat cosineWindow1D(cv::Mat input, int window_size = 16) {
+        cv::Mat cosineWindow1D(cv::Mat input, int window_size = TILE_SIZE) {
             cv::Mat output = input.clone();
             for (int i = 0; i < input.cols; ++i) {
                 output.at<float>(0, i) = 1. / 2. - 1. / 2. * cos(2 * M_PI * (input.at<float>(0, i) + 1 / 2.) / window_size);
@@ -55,7 +58,21 @@ class merge
             return window_applied;
         }
 
+        cv::Mat cat2Dtiles(std::vector<std::vector<cv::Mat>> tiles) {
+            std::vector<cv::Mat> rows;
+            for (auto row_tiles : tiles) {
+                cv::Mat row;
+                cv::hconcat(row_tiles, row);
+                rows.push_back(row);
+            }
+            cv::Mat img;
+            cv::vconcat(rows, img);
+            return img;
+        }
+
         std::vector<cv::Mat> getReferenceTiles(cv::Mat reference_image);
+
+        cv::Mat mergeTiles(std::vector<cv::Mat> tiles, int rows, int cols);
 };
 
 } // namespace hdrplus
diff --git a/src/merge.cpp b/src/merge.cpp
index ed7602b..0d20a53 100644
--- a/src/merge.cpp
+++ b/src/merge.cpp
@@ -13,27 +13,95 @@ void merge::process( const hdrplus::burst& burst_images, \
     double lambda_shot, lambda_read;
     std::tie(lambda_shot, lambda_read) = burst_images.bayer_images[burst_images.reference_image_idx].get_noise_params();
 
-    // Obtain tiles
-    cv::Mat reference_image = burst_images.grayscale_images_pad[0];
+    // Obtain Reference tiles
+    cv::Mat reference_image = burst_images.grayscale_images_pad[burst_images.reference_image_idx];
     std::vector<cv::Mat> reference_tiles = getReferenceTiles(reference_image);
 
+    // Temporal Denoising
+
+    // Spatial Denoising
+
+    // Process tiles through 2D cosine window
+    std::vector<cv::Mat> windowed_tiles;
+    for (auto tile : reference_tiles) {
+        windowed_tiles.push_back(cosineWindow2D(tile));
+    }
+
+    // Merge tiles
+    cv::Mat merged = mergeTiles(windowed_tiles, reference_image.rows, reference_image.cols);
+
     // cv::Mat outputImg = reference_image.clone();
     // cv::cvtColor(outputImg, outputImg, cv::COLOR_GRAY2RGB);
     // cv::imwrite("ref.jpg", outputImg);
+    // cv::Mat outputImg1 = reference_tiles[0].clone();
     // cv::Mat outputImg1 = cosineWindow2D(reference_tiles[0].clone());
+    // cv::Mat outputImg1 = cat2Dtiles(tiles_2D);
     // cv::cvtColor(outputImg1, outputImg1, cv::COLOR_GRAY2RGB);
     // cv::imwrite("tile0.jpg", outputImg1);
 }
 
 std::vector<cv::Mat> merge::getReferenceTiles(cv::Mat reference_image) {
     std::vector<cv::Mat> reference_tiles;
-    for (int y = 0; y < reference_image.rows - 8; y += 8) {
-        for (int x = 0; x < reference_image.cols - 8; x += 8) {
-            cv::Mat tile = reference_image(cv::Rect(x, y, 16, 16));
+    for (int y = 0; y < reference_image.rows - offset; y += offset) {
+        for (int x = 0; x < reference_image.cols - offset; x += offset) {
+            cv::Mat tile = reference_image(cv::Rect(x, y, TILE_SIZE, TILE_SIZE));
             reference_tiles.push_back(tile);
         }
     }
     return reference_tiles;
 }
 
+cv::Mat merge::mergeTiles(std::vector<cv::Mat> tiles, int num_rows, int num_cols){
+    // 1. get all four subsets: original (evenly split), horizontal overlapped,
+    // vertical overlapped, 2D overlapped
+    std::vector<std::vector<cv::Mat>> tiles_original;
+    for (int y = 0; y < num_rows / offset - 1; y += 2) {
+        std::vector<cv::Mat> row;
+        for (int x = 0; x < num_cols / offset - 1; x += 2) {
+            row.push_back(tiles[y * (num_cols / offset - 1) + x]);
+        }
+        tiles_original.push_back(row);
+    }
+
+    std::vector<std::vector<cv::Mat>> tiles_horizontal;
+    for (int y = 0; y < num_rows / offset - 1; y += 2) {
+        std::vector<cv::Mat> row;
+        for (int x = 1; x < num_cols / offset - 1; x += 2) {
+            row.push_back(tiles[y * (num_cols / offset - 1) + x]);
+        }
+        tiles_horizontal.push_back(row);
+    }
+
+    std::vector<std::vector<cv::Mat>> tiles_vertical;
+    for (int y = 1; y < num_rows / offset - 1; y += 2) {
+        std::vector<cv::Mat> row;
+        for (int x = 0; x < num_cols / offset - 1; x += 2) {
+            row.push_back(tiles[y * (num_cols / offset - 1) + x]);
+        }
+        tiles_vertical.push_back(row);
+    }
+
+    std::vector<std::vector<cv::Mat>> tiles_2d;
+    for (int y = 1; y < num_rows / offset - 1; y += 2) {
+        std::vector<cv::Mat> row;
+        for (int x = 1; x < num_cols / offset - 1; x += 2) {
+            row.push_back(tiles[y * (num_cols / offset - 1) + x]);
+        }
+        tiles_2d.push_back(row);
+    }
+
+    // 2. Concatenate the four subsets
+    cv::Mat img_original = cat2Dtiles(tiles_original);
+    cv::Mat img_horizontal = cat2Dtiles(tiles_horizontal);
+    cv::Mat img_vertical = cat2Dtiles(tiles_vertical);
+    cv::Mat img_2d = cat2Dtiles(tiles_2d);
+
+    // 3. Add the four subsets together
+    img_original(cv::Rect(offset, 0, num_cols - TILE_SIZE, num_rows)) += img_horizontal;
+    img_original(cv::Rect(0, offset, num_cols, num_rows - TILE_SIZE)) += img_vertical;
+    img_original(cv::Rect(offset, offset, num_cols - TILE_SIZE, num_rows - TILE_SIZE)) += img_2d;
+
+    return img_original;
+}
+
 } // namespace hdrplus