This homework aims to blend an object or texture from a source image into a target image using Poisson blending. This method matches the gradient of the image rather than the direct pixel values, allowing for a more seamless integration of the source object into the target image without noticeable seams. The process involves solving a least squares problem to find new pixel intensity values that preserve the source region's gradient within the target image.
The toy example result is shown below.
My favorite blend example is shown as follows (source, target, blend result in order).
The process involves selecting the floor lamp as the source region and defining a corresponding target region in the room. By solving the Poisson equation, we adjust the floor lamp’s pixel values to match the gradients of the surrounding room with a quite natural transition. Special adjustments I did was to carefully select the blending boundary.
Another example is shown as follows (source, target, blend result in order).
One example that is not quite well is shown as follows (source, target, blend result in order).
This test to blend a seagull into a beach scene does not work well. The contrast in lighting and texture between two images kind of mismatch, results in an blend with abnormal color. The main difficulty lies in blending objects from vastly different environments. Discrepancies in lighting, shadow, texture, and color gradients can lead to unnatural results. A solution could be adjusting the source object's lighting and shadows to better match the target scene before blending.