Rich Model (SRM)
Definition
A high-dimensional feature set for steganalysis constructed from joint statistics of pixel prediction residuals computed with many different filter kernels and quantisation steps. The Spatial Rich Model (SRM) produces around 34,671 features and is trained with an ensemble classifier.
- Full name
- Spatial Rich Model
- Feature count
- Approximately 34,671
- Basis
- Pixel prediction residual co-occurrence statistics
- Classifier
- Ensemble classifier
Common questions
Why does SRM use so many features instead of a small set?+
Steganographic embedding disturbs pixel statistics in subtle, filter-dependent ways, so combining many residual filters and quantisation steps captures a broader range of these disturbances than any single feature type, improving detection across different embedding algorithms.
What is a practical limitation of the full SRM feature set?+
The high dimensionality demands an ensemble classifier and substantial computation, and the model can still miss embedding schemes specifically designed to minimise the statistical residuals SRM measures.
Related terms
- Adaptive Embedding
- A steganographic strategy that concentrates embedding changes in high-texture or high-noise image regions where they are perceptually and statistically harder to detect....
- Calibration
- A steganalysis technique that estimates the cover image statistics by cropping or decompressing and re-compressing the test image to produce a reference....
- Chi-Square Attack
- A specific steganalysis test for LSB substitution in images. It tests whether pairs of pixel values that are related by flipping the...
- Convolutional Steganalysis Network
- A deep neural network trained end-to-end to classify cover versus stego images. Architectures such as XuNet, SRNet, and Yedroudj-Net use learned high-pass...
- LSB Substitution
- Least-significant-bit substitution: the lowest-order bit of each sample value (pixel colour channel or audio sample) is overwritten with one bit of the...