My homework 0 submission for the course CMSC733 Computer Processing of Pictorial Information. In this repository, a simple of version of PB boundary detection algorithm has been implemented. The classical approaches like Canny and Sobel edge detectors measures discontinuities in the image intensities to detect edges. The PB algorithm considers texture and color information along with intensity, making it a better performing algorithm.
Filter banks are a set of filters, that are applied over an image to extract multiple features. In this homework, filter banks have been used to extract the texture properties. The following subsections will describe in brief the implementation of DoG filters, Leung-Malik filters and Gabor filters.
The filters described in the previous section are to detect the texture properties in an image. Since each filter bank has multiple filters and three such filter banks have been used, the result is a vector of filter responses. This vector of filter response associated with each pixel is encoding a texture property. Based on this filter response vectors, pixels with similar texture property were clustered together using K-mean algorithm(K= 64). The output of K-mean clustering is a Texton map(τ )
The image was clustered based on the brightness value for each pixel. The images were first converted to gray scale and K-mean algorithm(K=16) was used to get brightness maps.
The image consists of three color channals(RGB), describ- ing the color property at each pixel. The images have been clustered using the RGB value using K-mean algorithm(K=16) to obtain color maps.
To obtain Tg,Bg,Cg, we need to compute differences of values across different shapes and sizes. This can be achieved very efficiently by the use of Half-disc masks.
The outputs from Sobel and Canny edge detector are combined using weighted average method.
In the final step, the features from baseline methods(Canny and Sobel operator) were combined with the gradients of τ , β, ζ.
.
├── Code
| ├── Wrapper.py
├── Data
| ├── BSDS500
├── Results
| ├── Filters
| ├── Testron_map
| ├── Brightness_map
| ├── Color_map
| ├── T_g
| ├── B_g
| ├── C_g
| ├── pb_lite_output
-
Change the location to
{root_directory}/Phase1 -
Run the following command
python3 Code/Wrapper.py
In this section, a basic neural network and its modified version for classification on CIFAR10 dataset have been de- scribed. Later, a case study for ResNet, ResNext and DenseNet architecture was conducted. Refer report for more details.