a classifier based on the shape of orientation histograms (SOH) of input images and random forests

Program Information

Name: a classifier based on the shape of orientation histograms (SOH) of input images and random forests
Domain: Machine learning
Functionality: The classifier makes use of SOH feature vectors to classify artcode or non-artcode images
Input: $I$: an image to be classified
Output: $Pr$: the probability of being classified as "artcode"

Reference


Enhancing Supervised Classifications with Metamorphic Relations
https://doi.org/10.1145/3193977.3193978


MR Information

MR1------

Description:
Property: $\sum_{i=1}^{n}Pr(B_{s_{a}}^{i}) \geq \sum_{i=1}^{n}Pr(B_{s_{n}}^{i})$: where $n$ is the number of image blocks; $Pr()$ is the probability to be classified as Artcode by the original classifier; $B_{s_{a}}^{i}$ denotes the ith block of the Artcode image; $B_{s_{n}}^{i}$ denotes the ith block of the non-artcode image;
Source input: $B_{s_{a}}^{i}$: splitting the input artcode image uniformly into blocks
Source output: $Pr(B_{s_{a}}^{i})$: the probability of the blocks being classified as "artcode"
Follow-up input: $B_{s_{n}}^{i}$: splitting the input non-artcode image uniformly into blocks
Follow-up output: $Pr(B_{s_{n}}^{i})$: the probability of the blocks being classified as "artcode"
Input relation: $B_{s_{a}}^{i}$: splitting the input artcode image uniformly into blocks, while $B_{s_{n}}^{i}$: splitting the input non-artcode image uniformly into blocks
Output relation: $\sum_{i=1}^{n}Pr(B_{s_{a}}^{i}) \geq \sum_{i=1}^{n}Pr(B_{s_{n}}^{i})$: $n$ is the number of image blocks; $Pr()$ is the probability to be classified as Artcode by the original classifier; $B_{s_{a}}^{i}$ denotes the ith block of the Artcode image; $B_{s_{n}}^{i}$ denotes the ith block of the non-artcode image;
Pattern: asymmetry, replacement

MR2------

Description:
Property: $\sum_{i=1}^{m}Pr(B_{O_{a}}^{i}) \geq \sum_{i=1}^{m}Pr(B_{O_{n}}^{i})$: where $m$ is the number of image masks; $B_{O_{a}}^{i} = \cap(I_{a},M_{i})$ and $B_{O_{n}}^{i} = \cap(I_{n},M_{i})$ outputs the overlapped areas of Artcode and non-Artcode images $Ia$ and $In$ and the ith mask $Mi$; $Pr()$ is the probability to be classified as Artcode by the original classifier; $B_{O_{a}}^{i}$ denotes the ith block of the Artcode image; $B_{O_{n}}^{i}$ denotes the ith block of the non-artcode image;
Source input: $B_{O_{a}}^{i}$: splitting the input artcode image (blocks with overlapped areas are permitted) into blocks
Source output: $Pr(O_{O_{a}}^{i})$: the probability of the blocks being classified as "artcode"
Follow-up input: $B_{O_{n}}^{i}$: splitting the input non-artcode image (blocks with overlapped areas are permitted) into blocks
Follow-up output: $Pr(B_{O_{n}}^{i})$: the probability of the blocks being classified as "artcode"
Input relation: $B_{O_{a}}^{i}$: splitting the input artcode image (blocks with overlapped areas are permitted) into blocks, while $B_{O_{n}}^{i}$: splitting the input non-artcode image (blocks with overlapped areas are permitted) into blocks
Output relation: $\sum_{i=1}^{m}Pr(B_{O_{a}}^{i}) \geq \sum_{i=1}^{m}Pr(B_{O_{n}}^{i})$: $m$ is the number of image masks; $B_{O_{a}}^{i} = \cap(I_{a},M_{i})$ and $B_{O_{n}}^{i} = \cap(I_{n},M_{i})$ outputs the overlapped areas of Artcode and non-Artcode images $Ia$ and $In$ and the ith mask $Mi$; $Pr()$ is the probability to be classified as Artcode by the original classifier; $B_{O_{a}}^{i}$ denotes the ith block of the Artcode image; $B_{O_{n}}^{i}$ denotes the ith block of the non-artcode image;
Pattern: asymmetry, replacement

MR3------

Description:
Property: $Pr(B_{O_{a}}^{i}) \geq Pr(B_{S_{a}}^{i})$: $Pr()$ is the probability to be classified as Artcode by the original classifier; $B_{O_{a}}^{i}$ denotes the ith block of the artcode image; $B_{S_{n}}^{i}$ denotes the ith block of the artcode image;
Source input: $B_{O_{a}}^{i}$: splitting the input artcode image (blocks with overlapped areas are permitted) into blocks
Source output: $Pr(O_{O_{a}}^{i})$: the probability of the blocks being classified as "artcode"
Follow-up input: $B_{S_{a}}^{i}$: splitting the input artcode image uniformly into blocks
Follow-up output: $Pr(B_{S_{a}}^{i})$: the probability of the blocks being classified as "artcode"
Input relation: $B_{O_{a}}^{i}$: splitting the input artcode image (blocks with overlapped areas are permitted) into blocks, while $B_{S_{a}}^{i}$: splitting the input artcode image uniformly into blocks
Output relation: $Pr(B_{O_{a}}^{i}) \geq Pr(B_{S_{a}}^{i})$: $Pr()$ is the probability to be classified as Artcode by the original classifier; $B_{O_{a}}^{i}$ denotes the ith block of the artcode image; $B_{S_{n}}^{i}$ denotes the ith block of the artcode image;
Pattern: asymmetry, replacement
Insert title here