MP1

Part1: Hybrid Images

1.1 The hybrid image of panda and lion

• Giving the low pass filter with a picture of low frequent texture and high pass filter with a picture of high frequent texture tends to produce the perfect hybrid image. Here, the panda is mostly charactrized by low frequent texures, and the lion has more high frequent details.
• By setting the low pass sigma to be 14.3558, we could filter the high frequent detail of panda such that it could only be visible in far distance.
• By setting the high pass sigma to be half of the low pass sigma(7.1779), the high frequent details of lion are both preserved and softened such that it would be seen in close distance but fade away in far distance.

Below are the log magnitude of Fourier transform of the images.

1.2 Another hybrid image (panda and tiger)

• Here, panda is still put into low pass filter, and tiger with high frequent texure is putted into high pass filter. The low pass sigma is 12.3554, and the high pass sigma is 5.7592.
• Another observation is two input images' outlines should be quite same but not too similar in order to make the image look not so ambiguous.

1.3 Failure example 1 (tiger and panther)

• Both tigers and panthers are characterized by high frequent textures, and most of the characteristics of tiger are filtered(with low pass sigma 6.6686) such that we cannot recognize tiger in far distance.
• The high pass sigma(6.6686) is also not enough to soften the texure of panther as its frequency is too high.
• At last, tiger and panther seem to be too similar, so that it is hard to see the difference.

1.4 Failure example 2 (monkey and lion)

• Again, the frequency of details of two images are too simmilar, such that we cannot choose good cutoffs. Both the low pass and high pass sigma here is 16.2020.
• Monkey and lion have great differences in shape of their nose and eyes. Therefore, the information provided by the hybrid image tends to be ambiguous.

Part2: Image Enhancement

2.1 Contrast Enhancement (gamma correction)

Gamma correction is implemented with factor 1/2.2, and histogram equilizattion is achieved by remapping v channel of the image. Both methods produce images that looks clearer with more details and contrast.

2.2 Color Enhancement

By taking the HSV color space of original image and applying s=min(1.0,2.5*s) in range [0,1] for every pixel, the image looks more saturated and brighter.

Extra Points

3.1 Color Shift

For the second image, color shift is achieved by converting image into Lab format and increasing a value. The third image is achieved by decreasing b value.

3.2 Gaussian and Laplacian pyramids

3.3 Using color to enhance the effect of bybrid images (marker and pen)

Coloring the pen(high-frequency component) makes it clearer.

Coloring the marker(low-frequency component) does not make too much difference.