CVE-2026-119338

Tesseract is an open source OCR engine. Prior to 5.5.4, ReadAdaptedClass in src/classify/adaptive.cpp raw-reads an ADAPT_CLASS_STRUCT from a pre-adapted-template .a file without neutralizing TempProtos and Config pointer fields before any destructor path. When the legacy engine is selected, classify_use_pre_adapted_templates is enabled, and an attacker can write the output-basename .a file, the success or rejection path can pass file-controlled pointers to ~ADAPT_CLASS_STRUCT and operator delete, causing memory corruption and denial of service.

In Tesseract prior to 5.5.4, GENERIC_2D_ARRAY::DeSerializeSize in src/ccstruct/matrix.h reads matrix dimensions as signed int32_t values and compares them only with UINT16_MAX, allowing negative values to reach ResizeNoInit. Negative dimensions or a 65535 by 65535 product can create inconsistent allocation state and later out-of-bounds accesses while loading a crafted traineddata file, causing denial of service.

In Tesseract prior to 5.5.4, WeightMatrix::DeSerialize trusts a mode byte from a traineddata LSTM network and calls WeightMatrix::DeSerializeOld when kDoubleFlag is clear. In default FAST_FLOAT builds, DeSerializeOld invokes ASSERT_HOST instead of returning an error, so loading a crafted model through TessBaseAPI::Init terminates the process and causes denial of service.

Tesseract 5.5.3 and earlier contains a flaw in Plumbing::DeSerialize that accepts a zero-length stack for NT_SERIES, NT_PARALLEL, or NT_REVERSED layers in a crafted .traineddata model. During LSTMRecognizer initialization, CacheXScaleFactor dereferences stack_[0] on the empty vector and invokes a virtual method through an invalid Network pointer.

Tesseract version 5.5.3 and earlier contains a heap overflow vulnerability in Classify::ReadIntTemplates, which reads NumClassPruners, NumClasses, and NumProtoSets from a .traineddata file without validating them against MAX_* limits. Oversized values cause out-of-bounds pointer writes on the heap during classifier initialization.

Tesseract version 5.5.3 and earlier contains a heap overflow vulnerability in UNICHARSET::load_via_fgets, which trusts the declared unichar count as a loop bound and uses id as an unchecked index into the unichars vector. This can lead to out-of-bounds writes during initialization.

Tesseract version 5.5.3 and earlier contains a heap overflow vulnerability in GenericVector::read, which reads reserved and size_used_ fields from a .traineddata file without caps. A crafted file can set reserved to a small value and size_used_ to a large value, causing out-of-bounds writes of FontInfo structures.

In Tesseract version 5.5.3 and earlier, RecodedCharID::DeSerialize in src/ccutil/unicharcompress.h validates length_ but accepts negative code_ values from a crafted .traineddata recoder component. UnicharCompress::ComputeCodeRange can consequently produce code_range_ equal to zero, after which SetupDecoder indexes is_valid_start_ with the negative code on a size-zero vector, causing an out-of-bounds bit write and a crash or allocation failure. No fixed release is available.

Tesseract 5.5.3 and earlier fails to validate dimension consistency in NetworkIO::WriteTimeStepPart and NetworkIO::AddTimeStepPart. A crafted .traineddata file with an NT_LSTM layer can cause a heap out-of-bounds write during the first recognition step on the default LSTM engine. No fixed release is available.

Tesseract 5.5.3 and earlier does not validate the deserialized scalars ni_ and no_ against weight-matrix dimensions in FullyConnected::DeSerialize. A crafted .traineddata file with an NT_SOFTMAX layer can cause a heap out-of-bounds write and read on the default LSTM engine. No fixed release is available.