REPORTING

Evaluation And Result Information

The project report should describe the Abouted system as multisteganalysis using CNN and Transformer. Results are reported as probability, confidence, reliability, and supporting evidence scores rather than unsupported accuracy claims.

TFM
Image transformer
CNN
Audio feature model
24
Video frames
JSON
Evidence report
WHAT TO REPORT IN YOUR FYP

Report the method as a CNN-Transformer multisteganalysis engine: transformer image inference, CNN audio inference, spatial LSB checks, JPEG-frequency checks, residual texture support, and video temporal aggregation. Report audio results from your trained checkpoints/audio_best.pth checkpoint when it is loaded.

RUNTIME BACKENDS
MediaBackendInputOutput
ImageTransformer + forensic enginesgrayscale 256x256 plus RGB bit-plane and DCT/residual statisticsP(stego), engine scores, reliability
Video24-frame temporal ensemblesampled frames across durationmean, P90, max, support, temporal artifact score
AudioAudio CNN + forensic gatemel + PCM-LSB + residual + spectral/sample-pair statisticsneural score, forensic score, final P(stego)
LIMITATIONS TO STATE CLEARLY

No universal steganalysis model detects every hiding tool. Lossless image LSB, JPEG DCT-domain hiding, audio LSB, phase artifacts, and video frame-level hiding leave different traces. The system therefore reports multiple evidence channels and a reliability label so the result is more transparent.