Package 'handwriterRF'

Title: Handwriting Analysis with Random Forests
Description: Perform forensic handwriting analysis of two scanned handwritten documents. This package implements the statistical method described by Madeline Johnson and Danica Ommen (2021) <doi:10.1002/sam.11566>. Similarity measures and a random forest produce a score-based likelihood ratio that quantifies the strength of the evidence in favor of the documents being written by the same writer or different writers.
Authors: Iowa State University of Science and Technology on behalf of its Center for Statistics and Applications in Forensic Evidence [aut, cph, fnd], Stephanie Reinders [aut, cre]
Maintainer: Stephanie Reinders <[email protected]>
License: GPL (>= 3)
Version: 1.0.2
Built: 2024-11-04 03:33:12 UTC
Source: CRAN

Help Index

Calculate a Score-Based Likelihood Ratio

Description

Compares two handwriting samples scanned and saved a PNG images with the following steps:

  1. processDocument splits the writing in both samples into component shapes, or graphs.

  2. get_clusters_batch groups the graphs into clusters of similar shapes.

  3. get_cluster_fill_counts counts the number of graphs assigned to each cluster.

  4. get_cluster_fill_rates calculates the proportion of graphs assigned to each cluster. The cluster fill rates serve as a writer profile.

  5. A similarity score is calculated between the cluster fill rates of the two documents using a random forest trained with ranger.

  6. The similarity score is compared to reference distributions of same writer and different writer similarity scores. The result is a score-based likelihood ratio that conveys the strength of the evidence in favor of same writer or different writer. For more details, see Madeline Johnson and Danica Ommen (2021) <doi:10.1002/sam.11566>.

Usage

calculate_slr(
  sample1_path,
  sample2_path,
  rforest = random_forest,
  project_dir = NULL
)

Arguments

sample1_path

A file path to a handwriting sample saved in PNG file format.
sample2_path

A file path to a second handwriting sample saved in PNG file format.
rforest

Optional. A random forest trained with ranger. If rforest is not given, the data object random_forest is used.
project_dir

Optional. A path to a directory where helper files will be saved. If no project directory is specified, the helper files will be saved to tempdir() and deleted before the function terminates.

Value

A number

Examples

# Compare two samples from the same writer
sample1 <- system.file(file.path("extdata", "w0030_s01_pWOZ_r01.png"), package = "handwriterRF")
sample2 <- system.file(file.path("extdata", "w0030_s01_pWOZ_r02.png"), package = "handwriterRF")
calculate_slr(sample1, sample2)

# Compare samples from two writers
sample1 <- system.file(file.path("extdata", "w0030_s01_pWOZ_r01.png"), package = "handwriterRF")
sample2 <- system.file(file.path("extdata", "w0238_s01_pWOZ_r02.png"), package = "handwriterRF")
calculate_slr(sample1, sample2)

Cluster Fill Counts for 1200 CSAFE Handwriting Database Samples

Description

A dataset containing cluster fill counts for for 1,200 handwriting samples from the CSAFE Handwriting Database. The documents were split into graphs with process_batch_dir. The graphs were grouped into clusters with get_clusters_batch. The cluster fill counts were calculated with get_cluster_fill_counts.

Usage

cfc

Format

A data frame with 1200 rows and 41 variables:

docname

The file name of the handwriting sample. The file name includes the writer ID, the writing session, prompt, and repetition number of the handwriting sample. There are 1,200 handwriting samples.

writer

Writer ID. There are 100 distinct writer ID's. Each writer has 12 documents.

doc

A document code that records the writing session, prompt, and repetition number of the handwriting sample. There are 12 distinct document codes. Each writer has a writing sample for each of the 12 document codes.

1

The number of graphs in cluster 1

2

The number of graphs in cluster 2

3

The number of graphs in cluster 3

4

The number of graphs in cluster 4

5

The number of graphs in cluster 5

6

The number of graphs in cluster 6

7

The number of graphs in cluster 7

8

The number of graphs in cluster 8

9

The number of graphs in cluster 9

10

The number of graphs in cluster 10

11

The number of graphs in cluster 11

12

The number of graphs in cluster 12

13

The number of graphs in cluster 13

14

The number of graphs in cluster 14

15

The number of graphs in cluster 15

16

The number of graphs in cluster 16

17

The number of graphs in cluster 17

18

The number of graphs in cluster 18

19

The number of graphs in cluster 19

20

The number of graphs in cluster 20

21

The number of graphs in cluster 21

22

The number of graphs in cluster 22

23

The number of graphs in cluster 23

24

The number of graphs in cluster 24

25

The number of graphs in cluster 25

26

The number of graphs in cluster 26

27

The number of graphs in cluster 27

28

The number of graphs in cluster 28

29

The number of graphs in cluster 29

30

The number of graphs in cluster 30

31

The number of graphs in cluster 31

32

The number of graphs in cluster 32

33

The number of graphs in cluster 33

34

The number of graphs in cluster 34

35

The number of graphs in cluster 35

36

The number of graphs in cluster 36

37

The number of graphs in cluster 37

38

The number of graphs in cluster 38

39

The number of graphs in cluster 39

40

The number of graphs in cluster 40

Source

<https://forensicstats.org/handwritingdatabase/>

Cluster Fill Rates for 1200 CSAFE Handwriting Database Samples

Description

A dataset containing cluster fill rates for for 1,200 handwriting samples from the CSAFE Handwriting Database. The dataset was created by running get_cluster_fill_rates on the cluster fill counts data frame cfc. Cluster fill rates are the proportion of total graphs assigned to each cluster.

Usage

cfr

Format

A data frame with 1200 rows and 42 variables:

docname

file name of the handwriting sample

total_graphs

The total number of graphs in the handwriting sample

cluster1

The number of graphs in cluster 1

cluster2

The number of graphs in cluster 2

cluster3

The number of graphs in cluster 3

cluster4

The number of graphs in cluster 4

cluster5

The number of graphs in cluster 5

cluster6

The number of graphs in cluster 6

cluster7

The number of graphs in cluster 7

cluster8

The number of graphs in cluster 8

cluster9

The number of graphs in cluster 9

cluster10

The number of graphs in cluster 10

cluster11

The number of graphs in cluster 11

cluster12

The number of graphs in cluster 12

cluster13

The number of graphs in cluster 13

cluster14

The number of graphs in cluster 14

cluster15

The number of graphs in cluster 15

cluster16

The number of graphs in cluster 16

cluster17

The number of graphs in cluster 17

cluster18

The number of graphs in cluster 18

cluster19

The number of graphs in cluster 19

cluster20

The number of graphs in cluster 20

cluster21

The number of graphs in cluster 21

cluster22

The number of graphs in cluster 22

cluster23

The number of graphs in cluster 23

cluster24

The number of graphs in cluster 24

cluster25

The number of graphs in cluster 25

cluster26

The number of graphs in cluster 26

cluster27

The number of graphs in cluster 27

cluster28

The number of graphs in cluster 28

cluster29

The number of graphs in cluster 29

cluster30

The number of graphs in cluster 30

cluster31

The number of graphs in cluster 31

cluster32

The number of graphs in cluster 32

cluster33

The number of graphs in cluster 33

cluster34

The number of graphs in cluster 34

cluster35

The number of graphs in cluster 35

cluster36

The number of graphs in cluster 36

cluster37

The number of graphs in cluster 37

cluster38

The number of graphs in cluster 38

cluster39

The number of graphs in cluster 39

cluster40

The number of graphs in cluster 40

Source

<https://forensicstats.org/handwritingdatabase/>

Get Cluster Fill Rates

Description

Calculate cluster fill rates from a data frame of cluster fill counts created with get_cluster_fill_counts.

Usage

get_cluster_fill_rates(df)

Arguments

df

A data frame of cluster fill rates created with get_cluster_fill_counts.

Value

A data frame of cluster fill rates.

Examples

rates <- get_cluster_fill_rates(df = cfc)

Get Training Set

Description

Create a training set from a data frame of cluster fill rates from the CSAFE Handwriting Database.

Usage

get_csafe_train_set(df, train_prompt_codes)

Arguments

df

A data frame of cluster fill rates created with get_cluster_fill_rates
train_prompt_codes

A character vector of which prompt(s) to use in the training set. Available prompts are 'pLND', 'pPHR', 'pWOZ', and 'pCMB'.

Value

A data frame

Examples

train <- get_csafe_train_set(df = cfr, train_prompt_codes = 'pCMB')

Get Distances

Description

Calculate distances using between all pairs of cluster fill rates in a data frame using one or more distance measures. The available distance measures absolute distance, Manhattan distance, Euclidean distance, maximum distance, and cosine distance.

Usage

get_distances(df, distance_measures)

Arguments

df

A data frame of cluster fill rates created with get_cluster_fill_rates
distance_measures

A vector of distance measures. Use 'abs' to calculate the absolute difference, 'man' for the Manhattan distance, 'euc' for the Euclidean distance, 'max' for the maximum absolute distance, and 'cos' for the cosine distance. The vector can be a single distance, or any combination of these five distance measures.

Details

The absolute distance between two n-length vectors of cluster fill rates, a and b, is a vector of the same length as a and b. It can be calculated as abs(a-b) where subtraction is performed element-wise, then the absolute value of each element is returned. More specifically, element i of the vector is ∣ai−bi∣|a_i - b_i| for i=1,2,...,ni=1,2,...,n.

The Manhattan distance between two n-length vectors of cluster fill rates, a and b, is ∑i=1n∣ai−bi∣\sum_{i=1}^n |a_i - b_i|. In other words, it is the sum of the absolute distance vector.

The Euclidean distance between two n-length vectors of cluster fill rates, a and b, is ∑i=1n(ai−bi)2\sqrt{\sum_{i=1}^n (a_i - b_i)^2}. In other words, it is the sum of the elements of the absolute distance vector.

The maximum distance between two n-length vectors of cluster fill rates, a and b, is max⁡1≤i≤n{∣ai−bi∣}\max_{1 \leq i \leq n}{\{|a_i - b_i|\}}. In other words, it is the sum of the elements of the absolute distance vector.

The cosine distance between two n-length vectors of cluster fill rates, a and b, is ∑i=1n(ai−bi)2/(∑i=1nai2∑i=1nbi2)\sum_{i=1}^n (a_i - b_i)^2 / (\sqrt{\sum_{i=1}^n a_i^2}\sqrt{\sum_{i=1}^n b_i^2}).

Value

A data frame of distances

Examples

# calculate maximum and Euclidean distances between the first 3 documents in cfr.
distances <- get_distances(df = cfr[1:3, ], distance_measures = c('max', 'euc'))

distances <- get_distances(df = cfr, distance_measures = c('man'))

Interpret an SLR Value

Description

Verbally interprent an SLR value.

Usage

interpret_slr(df)

Arguments

df

A data frame created by calculate_slr.

Value

A string

Examples

df <- data.frame("score" = 5, "slr" = 20)
interpret_slr(df)

df <- data.frame("score" = 0.12, "slr" = 0.5)
interpret_slr(df)

df <- data.frame("score" = 1, "slr" = 1)
interpret_slr(df)

df <- data.frame("score" = 0, "slr" = 0)
interpret_slr(df)

A ranger Random Forest, Distances, and Densities

Description

A list that contains a trained random forest created with ranger, the data frame of distances used to train the random forest, and two densities obtained from the random forest.

Usage

random_forest

Format

A list with the following components:

dists

The data frame used to train the random forest. The data frame has 600 rows. Each row contains the absolute and Euclidean distances between the cluster fill rates of two handwriting samples. If both handwriting samples are from the same writer, the class is 'same'. If the handwriting samples are from different writers, the class is 'different'. There are 300 'same' distances and 300 'different' distances in the data frame.

rf

A random forest created with ranger with settings: importance = 'permutation', scale.permutation.importance = TRUE, and num.trees = 200.

densities

A similarity score was obtained for each pair of handwriting samples in the training data frame, dists, by calculating the proportion of decision trees that voted 'same' class for the pair. The 'same_writer' density was created by applying density to the similarity scores for the 300 same writer pairs in dists. Similarly, the 'diff_writer' density was created by applying the density function to the similarity scores for the 300 different writer pairs in dists. The default settings were used with density.

Examples

# view the random forest
random_forest$rf

# view the distances data frame
random_forest$dists

# plot the same writer density
plot(random_forest$densities$same_writer)

# plot the different writer density
plot(random_forest$densities$diff_writer)

Cluster Template with 40 Clusters

Description

A cluster template created by handwriter with 40 clusters. This template was created from 120 handwriting samples from the CSAFE Handwriting Database.

Usage

templateK40

Format

A list containing the contents of the cluster template.

centers_seed

An integer for the random number generator use to select the starting cluster centers for the K-Means algorithm.

cluster

A vector of cluster assignments for each graph used to create the cluster template. The clusters are numbered sequentially 1, 2,...,K.

centers

The final cluster centers produced by the K-Means algorithm.

K

The number of clusters in the template.

n

The number of training graphs to used to create the template.

docnames

A vector that lists the training document from which each graph originated.

writers

A vector that lists the writer of each graph.

iters

The maximum number of iterations for the K-means algorithm.

changes

A vector of the number of graphs that changed clusters on each iteration of the K-means algorithm.

outlierCutoff

A vector of the outlier cutoff values calculated on each iteration of the K-means algorithm.

stop_reason

The reason the K-means algorithm terminated.

wcd

The within cluster distances on the final iteration of the K-means algorithm. More specifically, the distance between each graph and the center of the cluster to which it was assigned on each iteration. The output of make_clustering_template' stores the within cluster distances on each iteration, but the previous iterations were removed here to reduce the file size.

wcss

A vector of the within-cluster sum of squares on each iteration of the K-means algorithm.

Details

handwriter splits handwriting samples into component shapes called graphs. The graphs are sorted into 40 clusters with a K-Means algorithm.

Examples

# view number of clusters
templateK40$K

# view number of iterations
templateK40$iters

# view cluster centers
templateK40$centers

Train a Random Forest

Description

Train a random forest with ranger from a data frame of cluster fill rates.

Usage

train_rf(
  df,
  ntrees,
  distance_measures,
  output_dir = NULL,
  run_number = 1,
  downsample = TRUE
)

Arguments

df

A data frame of cluster fill rates created with get_cluster_fill_rates
ntrees

An integer number of decision trees to use
distance_measures

A vector of distance measures. Any combination of 'abs', 'euc', 'man', 'max', and 'cos' may be used.
output_dir

A path to a directory where the random forest will be saved.
run_number

An integer used for both the set.seed function and to distinguish between different runs on the same input data frame.
downsample

Whether to downsample the number of different writer distances before training the random forest. If TRUE, the different writer distances will be randomly sampled, resulting in the same number of different writer and same writer pairs.

Value

A random forest

Examples

train <- get_csafe_train_set(df = cfr, train_prompt_code = 'pCMB')
rforest <- train_rf(
  df = train,
  ntrees = 200,
  distance_measures = c('euc'),
  run_number = 1,
  downsample = TRUE
)