feat(backend): perceptual hashing for images and video

Adds a 64-bit dHash perceptual hash (internal/imagehash, built on the existing
disintegration/imaging — no new dependency) and starts populating the long-unused
data.files.phash column:

- Upload sets phash inline for images (cheap, from the in-memory bytes).
- Replace recomputes it from new content for images and clears it for anything
  else, so a stale hash never survives a content swap.
- FileRepo.SetPHash sets/clears the hash (used by Replace and, later, the dedup
  backfill).
- DiskStorage.VideoFrameMiddle extracts a frame from the middle of a clip
  (ffprobe duration -> ffmpeg -ss duration/2), avoiding the shared-intro collision
  a fixed early offset causes. It is a concrete method, not part of the storage
  port: only the dedup CLI needs it, keeping ffmpeg off the upload path. Video
  phashes are therefore computed by that CLI, not at upload time.
- DUPLICATE_HASH_THRESHOLD config (default 10/64) for the later pair rescan.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-06-16 12:20:52 +03:00
parent 58cea88f52
commit 88849cc16b
7 changed files with 278 additions and 7 deletions
+25
View File
@@ -16,6 +16,7 @@ import (
"github.com/google/uuid"
"tanabata/backend/internal/domain"
"tanabata/backend/internal/imagehash"
"tanabata/backend/internal/port"
)
@@ -154,6 +155,17 @@ func (s *FileService) Upload(ctx context.Context, p UploadParams) (*domain.File,
}
exifData, exifDatetime := extractMetadata(data, origName, p.ContentDatetimeFallback)
// Compute a perceptual hash for images so duplicate detection can later match
// near-identical files. Best-effort: a decode failure just leaves phash unset
// (the dedup CLI backfills it). Video is hashed by that CLI, not inline, to keep
// ffmpeg off the upload path.
var phash *int64
if strings.HasPrefix(mime.Name, "image/") {
if h, ok := imagehash.FromBytes(data); ok {
phash = &h
}
}
// Resolve content datetime: explicit > metadata date > fallback (e.g. import mtime) > zero.
var contentDatetime time.Time
if p.ContentDatetime != nil {
@@ -187,6 +199,7 @@ func (s *FileService) Upload(ctx context.Context, p UploadParams) (*domain.File,
Notes: p.Notes,
Metadata: p.Metadata,
EXIF: exifData,
PHash: phash,
CreatorID: userID,
IsPublic: p.IsPublic,
}
@@ -453,6 +466,18 @@ func (s *FileService) Replace(ctx context.Context, id uuid.UUID, p UploadParams)
return nil, err
}
// Recompute the perceptual hash from the new content: images inline, anything
// else cleared to NULL so the old content's hash never lingers (the dedup CLI
// recomputes video). Best-effort, like on upload — phash is recomputable.
var phash *int64
if strings.HasPrefix(mime.Name, "image/") {
if h, ok := imagehash.FromBytes(data); ok {
phash = &h
}
}
_ = s.files.SetPHash(ctx, id, phash)
updated.PHash = phash
objType := fileObjectType
_ = s.audit.Log(ctx, "file_replace", &objType, &id, nil)
return updated, nil