Calculate the histogram of pixel values

This is my code for the stage1 part of calculating the histogram and returning the most common pixel value for the image. How can I improve it because this parallelization is not giving me better performance?

    #pragma omp parallel for private(t_y, p_x, p_y, tile_index, tile_offset, pixel_offset, pixel) schedule(dynamic, 1)
    for (t_x = 0; t_x < openMP_TILES_X; ++t_x) {
        for (t_y = 0; t_y < openMP_TILES_Y; ++t_y) {
            const unsigned int tile_index = (t_y * openMP_TILES_X + t_x);
            const unsigned int tile_offset = (t_y * openMP_TILES_X * TILE_SIZE * TILE_SIZE + t_x * TILE_SIZE);
            // For each pixel within the tile
            for (int p_x = 0; p_x < TILE_SIZE; ++p_x) {
                for (int p_y = 0; p_y < TILE_SIZE; ++p_y) {
                    // Load pixel
                    const unsigned int pixel_offset = (p_y * openMP_input_image.width + p_x);
                    const unsigned char pixel = openMP_input_image.data[tile_offset + pixel_offset];
                    openMP_histograms[tile_index].histogram[pixel]++;
                    global_histogram[pixel]++;
                }
            }
        }
    }
    // Find the most common contrast value
    unsigned long long max_c = 0;  // Max count of pixels with a specific contrast value
    int max_i = -1; // Index (contrast value) of the histogram bin with max_c, init with an invalid value
#pragma omp parallel for private(i) schedule(dynamic)
    for (i = 0; i < PIXEL_RANGE; ++i) {
        if (max_c < global_histogram[i]) {
            max_c = global_histogram[i];
            max_i = i;
        }
    }